Add files using upload-large-folder tool
Browse files- .gitattributes +10 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00002-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00003-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00004-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00005-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00006-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00007-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00008-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00009-of-00010.gguf +3 -0
- IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00010-of-00010.gguf +3 -0
- IQ3_S/SHA256SUMS +10 -0
- IQ3_S/eval-offload.tsv +3 -0
- IQ3_S/eval-ppl.log +276 -0
- IQ3_S/eval-ppl.ngl99.log +276 -0
- IQ3_S/quantize-command.txt +8 -0
- IQ3_S/quantize.log +0 -0
- IQ3_S/result.tsv +2 -0
- IQ3_S/smoke.log +285 -0
- IQ3_S/smoke.ngl99.log +285 -0
- MANIFEST.tsv +21 -18
- README.md +1 -1
- results.tsv +1 -0
.gitattributes
CHANGED
|
@@ -53,3 +53,13 @@ IQ3_K_R4/Laguna-M.1-IQ3_K_R4-imatrix-public-v1-00004-of-00010.gguf filter=lfs di
|
|
| 53 |
IQ3_K_R4/Laguna-M.1-IQ3_K_R4-imatrix-public-v1-00008-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 54 |
IQ3_K_R4/Laguna-M.1-IQ3_K_R4-imatrix-public-v1-00005-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 55 |
IQ3_K_R4/Laguna-M.1-IQ3_K_R4-imatrix-public-v1-00003-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
IQ3_K_R4/Laguna-M.1-IQ3_K_R4-imatrix-public-v1-00008-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 54 |
IQ3_K_R4/Laguna-M.1-IQ3_K_R4-imatrix-public-v1-00005-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 55 |
IQ3_K_R4/Laguna-M.1-IQ3_K_R4-imatrix-public-v1-00003-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 56 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00002-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 57 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00006-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 58 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 59 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00003-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 60 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00004-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 61 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00007-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 62 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00008-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 63 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00010-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 64 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00009-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
| 65 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00005-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:95164a5592987dff50a9fad555b56c148b9d01ceff95816e2fa1175424151c60
|
| 3 |
+
size 11377052512
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00002-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4b18ba2aafa656734013a325bbd5858fd3eecb026705094c904a96692cf7ca03
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00003-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d1cd6f3e2501029ce8c8ce82b6bbc21c3fd05f571070a7c3a6aca0969b3d4c9f
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00004-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4cee36fc0918c7fcb4ba728ff6790117e1beba642cec9b1104f6fab5ba2a6000
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00005-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2aa0b85f00f95c64107ce6a5e847d6c19291da3d931bb509f2ade52578b82fd1
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00006-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0390cc1e252052fb5e07e221c97a5cbb8c779483bd4706f0325b3818f51c1d94
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00007-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:63e90c41ea981b62ace5b1edc84096a3ba51d55db2759f68eb59786f4127a48c
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00008-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:27b8f023b41e0cc7f7a73955b84f0877b7ce6d01f288a16c631dac43c5aa63ba
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00009-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8311a8576b1a8d6ace41fa9adee21f14f01622dcfd959e08a27f839534b22edf
|
| 3 |
+
size 10150217440
|
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00010-of-00010.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:649469aec90078da7b0e6263a3bed109769872c900e1f9176542d08e1d197647
|
| 3 |
+
size 5997855808
|
IQ3_S/SHA256SUMS
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
95164a5592987dff50a9fad555b56c148b9d01ceff95816e2fa1175424151c60 Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf
|
| 2 |
+
4b18ba2aafa656734013a325bbd5858fd3eecb026705094c904a96692cf7ca03 Laguna-M.1-IQ3_S-imatrix-public-v1-00002-of-00010.gguf
|
| 3 |
+
d1cd6f3e2501029ce8c8ce82b6bbc21c3fd05f571070a7c3a6aca0969b3d4c9f Laguna-M.1-IQ3_S-imatrix-public-v1-00003-of-00010.gguf
|
| 4 |
+
4cee36fc0918c7fcb4ba728ff6790117e1beba642cec9b1104f6fab5ba2a6000 Laguna-M.1-IQ3_S-imatrix-public-v1-00004-of-00010.gguf
|
| 5 |
+
2aa0b85f00f95c64107ce6a5e847d6c19291da3d931bb509f2ade52578b82fd1 Laguna-M.1-IQ3_S-imatrix-public-v1-00005-of-00010.gguf
|
| 6 |
+
0390cc1e252052fb5e07e221c97a5cbb8c779483bd4706f0325b3818f51c1d94 Laguna-M.1-IQ3_S-imatrix-public-v1-00006-of-00010.gguf
|
| 7 |
+
63e90c41ea981b62ace5b1edc84096a3ba51d55db2759f68eb59786f4127a48c Laguna-M.1-IQ3_S-imatrix-public-v1-00007-of-00010.gguf
|
| 8 |
+
27b8f023b41e0cc7f7a73955b84f0877b7ce6d01f288a16c631dac43c5aa63ba Laguna-M.1-IQ3_S-imatrix-public-v1-00008-of-00010.gguf
|
| 9 |
+
8311a8576b1a8d6ace41fa9adee21f14f01622dcfd959e08a27f839534b22edf Laguna-M.1-IQ3_S-imatrix-public-v1-00009-of-00010.gguf
|
| 10 |
+
649469aec90078da7b0e6263a3bed109769872c900e1f9176542d08e1d197647 Laguna-M.1-IQ3_S-imatrix-public-v1-00010-of-00010.gguf
|
IQ3_S/eval-offload.tsv
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
phase ngl status log
|
| 2 |
+
smoke 99 ok /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/smoke.ngl99.log
|
| 3 |
+
ppl 99 ok /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/eval-ppl.ngl99.log
|
IQ3_S/eval-ppl.log
ADDED
|
@@ -0,0 +1,276 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
main: build = 4641 (be7d53ce)
|
| 2 |
+
main: built with cc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0 for aarch64-linux-gnu
|
| 3 |
+
main: seed = 42
|
| 4 |
+
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
|
| 5 |
+
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
|
| 6 |
+
ggml_cuda_init: found 1 CUDA devices:
|
| 7 |
+
Device 0: NVIDIA GB10, compute capability 12.1, VMM: yes, VRAM: 124610 MiB
|
| 8 |
+
CUDA0: using device CUDA0 - 96489 MiB free
|
| 9 |
+
llama_model_loader: additional 9 GGUFs metadata loaded.
|
| 10 |
+
llama_model_loader: loaded meta data with 59 key-value pairs and 1178 tensors from /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf (version GGUF V3 (latest))
|
| 11 |
+
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
|
| 12 |
+
llama_model_loader: - kv 0: general.architecture str = laguna
|
| 13 |
+
llama_model_loader: - kv 1: general.type str = model
|
| 14 |
+
llama_model_loader: - kv 2: general.name str = Laguna M.1
|
| 15 |
+
llama_model_loader: - kv 3: general.size_label str = 256x11B
|
| 16 |
+
llama_model_loader: - kv 4: general.license str = apache-2.0
|
| 17 |
+
llama_model_loader: - kv 5: general.tags arr[str,6] = ["laguna-m.1", "vllm", "sglang", "bf1...
|
| 18 |
+
llama_model_loader: - kv 6: laguna.context_length u32 = 262144
|
| 19 |
+
llama_model_loader: - kv 7: laguna.embedding_length u32 = 4096
|
| 20 |
+
llama_model_loader: - kv 8: laguna.block_count u32 = 70
|
| 21 |
+
llama_model_loader: - kv 9: laguna.feed_forward_length u32 = 16384
|
| 22 |
+
llama_model_loader: - kv 10: laguna.attention.head_count arr[i32,70] = [64, 64, 64, 64, 64, 64, 64, 64, 64, ...
|
| 23 |
+
llama_model_loader: - kv 11: laguna.attention.head_count_kv u32 = 8
|
| 24 |
+
llama_model_loader: - kv 12: laguna.attention.key_length u32 = 128
|
| 25 |
+
llama_model_loader: - kv 13: laguna.attention.value_length u32 = 128
|
| 26 |
+
llama_model_loader: - kv 14: laguna.attention.layer_norm_rms_epsilon f32 = 0.000001
|
| 27 |
+
llama_model_loader: - kv 15: general.file_type u32 = 26
|
| 28 |
+
llama_model_loader: - kv 16: laguna.attention.sliding_window u32 = 0
|
| 29 |
+
llama_model_loader: - kv 17: laguna.rope.dimension_count u32 = 128
|
| 30 |
+
llama_model_loader: - kv 18: laguna.rope.dimension_count_swa u32 = 128
|
| 31 |
+
llama_model_loader: - kv 19: laguna.rope.freq_base f32 = 500000.000000
|
| 32 |
+
llama_model_loader: - kv 20: laguna.rope.freq_base_swa f32 = 10000.000000
|
| 33 |
+
llama_model_loader: - kv 21: laguna.rope.scaling.type str = yarn
|
| 34 |
+
llama_model_loader: - kv 22: laguna.rope.scaling.factor f32 = 64.000000
|
| 35 |
+
llama_model_loader: - kv 23: laguna.rope.scaling.original_context_length u32 = 4096
|
| 36 |
+
llama_model_loader: - kv 24: laguna.rope.scaling.yarn_ext_factor f32 = 1.000000
|
| 37 |
+
llama_model_loader: - kv 25: laguna.rope.scaling.yarn_attn_factor f32 = 1.000000
|
| 38 |
+
llama_model_loader: - kv 26: laguna.rope.scaling.yarn_beta_fast f32 = 64.000000
|
| 39 |
+
llama_model_loader: - kv 27: laguna.rope.scaling.yarn_beta_slow f32 = 1.000000
|
| 40 |
+
llama_model_loader: - kv 28: laguna.expert_count u32 = 256
|
| 41 |
+
llama_model_loader: - kv 29: laguna.expert_used_count u32 = 16
|
| 42 |
+
llama_model_loader: - kv 30: laguna.expert_feed_forward_length u32 = 1024
|
| 43 |
+
llama_model_loader: - kv 31: laguna.expert_shared_feed_forward_length u32 = 1024
|
| 44 |
+
llama_model_loader: - kv 32: laguna.expert_weights_scale f32 = 1.000000
|
| 45 |
+
llama_model_loader: - kv 33: laguna.expert_weights_norm bool = true
|
| 46 |
+
llama_model_loader: - kv 34: laguna.expert_gating_func u32 = 2
|
| 47 |
+
llama_model_loader: - kv 35: laguna.leading_dense_block_count u32 = 3
|
| 48 |
+
llama_model_loader: - kv 36: tokenizer.ggml.model str = gpt2
|
| 49 |
+
llama_model_loader: - kv 37: tokenizer.ggml.pre str = laguna
|
| 50 |
+
llama_model_loader: - kv 38: tokenizer.ggml.tokens arr[str,100352] = ["γ|UNK|γ", "γ|CODE_START|γ",...
|
| 51 |
+
llama_model_loader: - kv 39: tokenizer.ggml.token_type arr[i32,100352] = [3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, ...
|
| 52 |
+
llama_model_loader: - kv 40: tokenizer.ggml.merges arr[str,100026] = ["i n", "Δ t", "Δ Δ ", "e r", "Δ a...
|
| 53 |
+
llama_model_loader: - kv 41: tokenizer.ggml.bos_token_id u32 = 2
|
| 54 |
+
llama_model_loader: - kv 42: tokenizer.ggml.eos_token_id u32 = 2
|
| 55 |
+
llama_model_loader: - kv 43: tokenizer.ggml.unknown_token_id u32 = 0
|
| 56 |
+
llama_model_loader: - kv 44: tokenizer.ggml.seperator_token_id u32 = 8
|
| 57 |
+
llama_model_loader: - kv 45: tokenizer.ggml.padding_token_id u32 = 9
|
| 58 |
+
llama_model_loader: - kv 46: tokenizer.ggml.mask_token_id u32 = 12
|
| 59 |
+
llama_model_loader: - kv 47: tokenizer.ggml.add_bos_token bool = true
|
| 60 |
+
llama_model_loader: - kv 48: tokenizer.ggml.add_sep_token bool = true
|
| 61 |
+
llama_model_loader: - kv 49: tokenizer.chat_template str = {#- Copied from laguna_glm_thinking_v...
|
| 62 |
+
llama_model_loader: - kv 50: tokenizer.ggml.eot_token_id u32 = 24
|
| 63 |
+
llama_model_loader: - kv 51: general.quantization_version u32 = 2
|
| 64 |
+
llama_model_loader: - kv 52: quantize.imatrix.file str = /mnt/pool/gguf/laguna-m1/imatrix/q8-p...
|
| 65 |
+
llama_model_loader: - kv 53: quantize.imatrix.dataset str = /workspace/laguna-m1/corpus/laguna-m1...
|
| 66 |
+
llama_model_loader: - kv 54: quantize.imatrix.entries_count i32 = 828
|
| 67 |
+
llama_model_loader: - kv 55: quantize.imatrix.chunks_count i32 = 2048
|
| 68 |
+
llama_model_loader: - kv 56: split.no u16 = 0
|
| 69 |
+
llama_model_loader: - kv 57: split.count u16 = 10
|
| 70 |
+
llama_model_loader: - kv 58: split.tensors.count i32 = 1178
|
| 71 |
+
llama_model_loader: - type f32: 415 tensors
|
| 72 |
+
llama_model_loader: - type q8_0: 140 tensors
|
| 73 |
+
llama_model_loader: - type q5_K: 70 tensors
|
| 74 |
+
llama_model_loader: - type q6_K: 2 tensors
|
| 75 |
+
llama_model_loader: - type iq3_s: 551 tensors
|
| 76 |
+
load: 0 unused tokens
|
| 77 |
+
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 78 |
+
load: special_eot_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 79 |
+
load: printing all EOG tokens:
|
| 80 |
+
load: - 2 ('γ|EOS|γ')
|
| 81 |
+
load: - 24 ('</assistant>')
|
| 82 |
+
load: special tokens cache size = 70
|
| 83 |
+
load: token to piece cache size = 0.6432 MB
|
| 84 |
+
llm_load_print_meta: format = GGUF V3 (latest)
|
| 85 |
+
llm_load_print_meta: arch = laguna
|
| 86 |
+
llm_load_print_meta: n_ctx_train = 262144
|
| 87 |
+
llm_load_print_meta: n_embd = 4096
|
| 88 |
+
llm_load_print_meta: n_layer = 70
|
| 89 |
+
llm_load_print_meta: n_head = 64
|
| 90 |
+
llm_load_print_meta: n_head_kv = 8
|
| 91 |
+
llm_load_print_meta: n_rot = 128
|
| 92 |
+
llm_load_print_meta: n_swa = 0
|
| 93 |
+
llm_load_print_meta: n_swa_pattern = 1
|
| 94 |
+
llm_load_print_meta: n_embd_head_k = 128
|
| 95 |
+
llm_load_print_meta: n_embd_head_v = 128
|
| 96 |
+
llm_load_print_meta: n_gqa = 8
|
| 97 |
+
llm_load_print_meta: n_embd_k_gqa = 1024
|
| 98 |
+
llm_load_print_meta: n_embd_v_gqa = 1024
|
| 99 |
+
llm_load_print_meta: f_norm_eps = 0.0e+00
|
| 100 |
+
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
|
| 101 |
+
llm_load_print_meta: f_clamp_kqv = 0.0e+00
|
| 102 |
+
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
|
| 103 |
+
llm_load_print_meta: f_logit_scale = 0.0e+00
|
| 104 |
+
llm_load_print_meta: n_ff = 16384
|
| 105 |
+
llm_load_print_meta: n_expert = 256
|
| 106 |
+
llm_load_print_meta: n_expert_used = 16
|
| 107 |
+
llm_load_print_meta: causal attn = 1
|
| 108 |
+
llm_load_print_meta: pooling type = 0
|
| 109 |
+
llm_load_print_meta: rope type = 2
|
| 110 |
+
llm_load_print_meta: rope scaling = yarn
|
| 111 |
+
llm_load_print_meta: freq_base_train = 500000.0
|
| 112 |
+
llm_load_print_meta: freq_scale_train = 0.015625
|
| 113 |
+
llm_load_print_meta: n_ctx_orig_yarn = 4096
|
| 114 |
+
llm_load_print_meta: rope_finetuned = unknown
|
| 115 |
+
llm_load_print_meta: ssm_d_conv = 0
|
| 116 |
+
llm_load_print_meta: ssm_d_inner = 0
|
| 117 |
+
llm_load_print_meta: ssm_d_state = 0
|
| 118 |
+
llm_load_print_meta: ssm_dt_rank = 0
|
| 119 |
+
llm_load_print_meta: ssm_n_group = 0
|
| 120 |
+
llm_load_print_meta: model type = ?B
|
| 121 |
+
llm_load_print_meta: model ftype = IQ3_S - 3.4375 bpw
|
| 122 |
+
llm_load_print_meta: model params = 225.796 B
|
| 123 |
+
llm_load_print_meta: model size = 91.803 GiB (3.492 BPW)
|
| 124 |
+
llm_load_print_meta: repeating layers = 91.175 GiB (3.481 BPW, 224.974 B parameters)
|
| 125 |
+
llm_load_print_meta: general.name = Laguna M.1
|
| 126 |
+
print_info: vocab type = BPE
|
| 127 |
+
print_info: n_vocab = 100352
|
| 128 |
+
print_info: n_merges = 100026
|
| 129 |
+
print_info: BOS token = 2 'γ|EOS|γ'
|
| 130 |
+
print_info: EOS token = 2 'γ|EOS|γ'
|
| 131 |
+
print_info: EOT token = 24 '</assistant>'
|
| 132 |
+
print_info: UNK token = 0 'γ|UNK|γ'
|
| 133 |
+
print_info: SEP token = 8 'γ|SEP|γ'
|
| 134 |
+
print_info: PAD token = 9 'γ|PAD|γ'
|
| 135 |
+
print_info: MASK token = 12 'γ|MASK|γ'
|
| 136 |
+
print_info: LF token = 268 'Δ'
|
| 137 |
+
print_info: EOG token = 2 'γ|EOS|γ'
|
| 138 |
+
print_info: EOG token = 24 '</assistant>'
|
| 139 |
+
print_info: max token length = 830
|
| 140 |
+
======================================= HAVE_FANCY_SIMD is NOT defined
|
| 141 |
+
------------------- Layer sizes:
|
| 142 |
+
Layer 0: 140.53, 16.00, 156.53 22.00 MiB
|
| 143 |
+
Layer 1: 140.53, 16.00, 156.53 22.00 MiB
|
| 144 |
+
Layer 2: 140.53, 16.00, 156.53 22.00 MiB
|
| 145 |
+
Layer 3: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 146 |
+
Layer 4: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 147 |
+
Layer 5: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 148 |
+
Layer 6: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 149 |
+
Layer 7: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 150 |
+
Layer 8: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 151 |
+
Layer 9: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 152 |
+
Layer 10: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 153 |
+
Layer 11: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 154 |
+
Layer 12: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 155 |
+
Layer 13: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 156 |
+
Layer 14: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 157 |
+
Layer 15: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 158 |
+
Layer 16: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 159 |
+
Layer 17: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 160 |
+
Layer 18: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 161 |
+
Layer 19: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 162 |
+
Layer 20: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 163 |
+
Layer 21: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 164 |
+
Layer 22: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 165 |
+
Layer 23: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 166 |
+
Layer 24: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 167 |
+
Layer 25: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 168 |
+
Layer 26: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 169 |
+
Layer 27: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 170 |
+
Layer 28: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 171 |
+
Layer 29: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 172 |
+
Layer 30: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 173 |
+
Layer 31: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 174 |
+
Layer 32: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 175 |
+
Layer 33: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 176 |
+
Layer 34: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 177 |
+
Layer 35: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 178 |
+
Layer 36: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 179 |
+
Layer 37: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 180 |
+
Layer 38: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 181 |
+
Layer 39: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 182 |
+
Layer 40: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 183 |
+
Layer 41: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 184 |
+
Layer 42: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 185 |
+
Layer 43: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 186 |
+
Layer 44: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 187 |
+
Layer 45: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 188 |
+
Layer 46: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 189 |
+
Layer 47: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 190 |
+
Layer 48: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 191 |
+
Layer 49: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 192 |
+
Layer 50: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 193 |
+
Layer 51: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 194 |
+
Layer 52: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 195 |
+
Layer 53: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 196 |
+
Layer 54: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 197 |
+
Layer 55: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 198 |
+
Layer 56: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 199 |
+
Layer 57: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 200 |
+
Layer 58: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 201 |
+
Layer 59: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 202 |
+
Layer 60: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 203 |
+
Layer 61: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 204 |
+
Layer 62: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 205 |
+
Layer 63: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 206 |
+
Layer 64: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 207 |
+
Layer 65: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 208 |
+
Layer 66: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 209 |
+
Layer 67: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 210 |
+
Layer 68: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 211 |
+
Layer 69: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 212 |
+
Layer 70: 321.56, 0.00, 321.56 MiB (output layer)
|
| 213 |
+
--------------------------------------------------------------------------
|
| 214 |
+
Total : 93363.29, 1120.00, 94483.29 MiB
|
| 215 |
+
Memory required for model tensors + cache: 94805 MiB
|
| 216 |
+
Memory available on all devices - compute: 95379 MiB
|
| 217 |
+
llm_load_tensors: ggml ctx size = 1.02 MiB
|
| 218 |
+
llm_load_tensors: offloading 70 repeating layers to GPU
|
| 219 |
+
llm_load_tensors: offloading non-repeating layers to GPU
|
| 220 |
+
llm_load_tensors: offloaded 71/71 layers to GPU
|
| 221 |
+
llm_load_tensors: CPU buffer size = 321.56 MiB
|
| 222 |
+
llm_load_tensors: CUDA0 buffer size = 93684.87 MiB
|
| 223 |
+
....................................................................................................
|
| 224 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 225 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 226 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 227 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 228 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 229 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 230 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 231 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 232 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 233 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 234 |
+
llm_load_tensors: dense parameters loaded in 110.62s (5.44 GiB), expert parameters deferred (86.37 GiB)
|
| 235 |
+
llama_init_from_model: n_ctx = 4096
|
| 236 |
+
llama_init_from_model: n_batch = 512
|
| 237 |
+
llama_init_from_model: n_ubatch = 128
|
| 238 |
+
llama_init_from_model: flash_attn = 1
|
| 239 |
+
llama_init_from_model: attn_max_b = 0
|
| 240 |
+
llama_init_from_model: fused_moe = 1
|
| 241 |
+
llama_init_from_model: grouped er = 0
|
| 242 |
+
llama_init_from_model: fused_up_gate = 1
|
| 243 |
+
llama_init_from_model: fused_mmad = 1
|
| 244 |
+
llama_init_from_model: rope_cache = 0
|
| 245 |
+
llama_init_from_model: graph_reuse = 1
|
| 246 |
+
llama_init_from_model: k_cache_hadam = 0
|
| 247 |
+
llama_init_from_model: v_cache_hadam = 0
|
| 248 |
+
llama_init_from_model: split_mode_graph_scheduling = 0
|
| 249 |
+
llama_init_from_model: reduce_type = f16
|
| 250 |
+
llama_init_from_model: sched_async = 0
|
| 251 |
+
llama_init_from_model: ser = -1, 0
|
| 252 |
+
llama_init_from_model: freq_base = 500000.0
|
| 253 |
+
llama_init_from_model: freq_scale = 0.015625
|
| 254 |
+
llama_kv_cache_init: CUDA0 KV buffer size = 1120.00 MiB
|
| 255 |
+
llama_init_from_model: KV self size = 1120.00 MiB, K (f16): 560.00 MiB, V (f16): 560.00 MiB
|
| 256 |
+
llama_init_from_model: CUDA_Host output buffer size = 0.38 MiB
|
| 257 |
+
llama_init_from_model: CUDA0 compute buffer size = 51.00 MiB
|
| 258 |
+
llama_init_from_model: CUDA_Host compute buffer size = 3.00 MiB
|
| 259 |
+
llama_init_from_model: graph nodes = 3174
|
| 260 |
+
llama_init_from_model: graph splits = 2
|
| 261 |
+
llama_init_from_model: enabling only_active_experts scheduling
|
| 262 |
+
|
| 263 |
+
system_info: n_threads = 20 / 20 | AVX = 0 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 0 | NEON = 1 | SVE = 0 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | MATMUL_INT8 = 0 |
|
| 264 |
+
perplexity: tokenizing the input ..
|
| 265 |
+
perplexity: tokenization took 1789.63 ms
|
| 266 |
+
perplexity: calculating perplexity over 8 chunks, n_ctx=4096, batch_size=512, n_seq=1
|
| 267 |
+
perplexity: 25.98 seconds per pass - ETA 3.45 minutes
|
| 268 |
+
[1]1.3713,[2]2.2535,[3]1.8046,[4]2.4539,[5]2.6550,[6]2.6851,[7]2.6250,[8]2.4683,
|
| 269 |
+
llama_print_timings: load time = 128307.95 ms
|
| 270 |
+
llama_print_timings: sample time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
| 271 |
+
llama_print_timings: prompt eval time = 209390.13 ms / 32768 tokens ( 6.39 ms per token, 156.49 tokens per second)
|
| 272 |
+
llama_print_timings: eval time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
| 273 |
+
llama_print_timings: total time = 322931.50 ms / 32769 tokens
|
| 274 |
+
~ggml_backend_cuda_context: have 1 graphs
|
| 275 |
+
|
| 276 |
+
Final estimate: PPL over 8 chunks for n_ctx=4096 = 2.4683 +/- 0.04051
|
IQ3_S/eval-ppl.ngl99.log
ADDED
|
@@ -0,0 +1,276 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
main: build = 4641 (be7d53ce)
|
| 2 |
+
main: built with cc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0 for aarch64-linux-gnu
|
| 3 |
+
main: seed = 42
|
| 4 |
+
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
|
| 5 |
+
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
|
| 6 |
+
ggml_cuda_init: found 1 CUDA devices:
|
| 7 |
+
Device 0: NVIDIA GB10, compute capability 12.1, VMM: yes, VRAM: 124610 MiB
|
| 8 |
+
CUDA0: using device CUDA0 - 96489 MiB free
|
| 9 |
+
llama_model_loader: additional 9 GGUFs metadata loaded.
|
| 10 |
+
llama_model_loader: loaded meta data with 59 key-value pairs and 1178 tensors from /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf (version GGUF V3 (latest))
|
| 11 |
+
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
|
| 12 |
+
llama_model_loader: - kv 0: general.architecture str = laguna
|
| 13 |
+
llama_model_loader: - kv 1: general.type str = model
|
| 14 |
+
llama_model_loader: - kv 2: general.name str = Laguna M.1
|
| 15 |
+
llama_model_loader: - kv 3: general.size_label str = 256x11B
|
| 16 |
+
llama_model_loader: - kv 4: general.license str = apache-2.0
|
| 17 |
+
llama_model_loader: - kv 5: general.tags arr[str,6] = ["laguna-m.1", "vllm", "sglang", "bf1...
|
| 18 |
+
llama_model_loader: - kv 6: laguna.context_length u32 = 262144
|
| 19 |
+
llama_model_loader: - kv 7: laguna.embedding_length u32 = 4096
|
| 20 |
+
llama_model_loader: - kv 8: laguna.block_count u32 = 70
|
| 21 |
+
llama_model_loader: - kv 9: laguna.feed_forward_length u32 = 16384
|
| 22 |
+
llama_model_loader: - kv 10: laguna.attention.head_count arr[i32,70] = [64, 64, 64, 64, 64, 64, 64, 64, 64, ...
|
| 23 |
+
llama_model_loader: - kv 11: laguna.attention.head_count_kv u32 = 8
|
| 24 |
+
llama_model_loader: - kv 12: laguna.attention.key_length u32 = 128
|
| 25 |
+
llama_model_loader: - kv 13: laguna.attention.value_length u32 = 128
|
| 26 |
+
llama_model_loader: - kv 14: laguna.attention.layer_norm_rms_epsilon f32 = 0.000001
|
| 27 |
+
llama_model_loader: - kv 15: general.file_type u32 = 26
|
| 28 |
+
llama_model_loader: - kv 16: laguna.attention.sliding_window u32 = 0
|
| 29 |
+
llama_model_loader: - kv 17: laguna.rope.dimension_count u32 = 128
|
| 30 |
+
llama_model_loader: - kv 18: laguna.rope.dimension_count_swa u32 = 128
|
| 31 |
+
llama_model_loader: - kv 19: laguna.rope.freq_base f32 = 500000.000000
|
| 32 |
+
llama_model_loader: - kv 20: laguna.rope.freq_base_swa f32 = 10000.000000
|
| 33 |
+
llama_model_loader: - kv 21: laguna.rope.scaling.type str = yarn
|
| 34 |
+
llama_model_loader: - kv 22: laguna.rope.scaling.factor f32 = 64.000000
|
| 35 |
+
llama_model_loader: - kv 23: laguna.rope.scaling.original_context_length u32 = 4096
|
| 36 |
+
llama_model_loader: - kv 24: laguna.rope.scaling.yarn_ext_factor f32 = 1.000000
|
| 37 |
+
llama_model_loader: - kv 25: laguna.rope.scaling.yarn_attn_factor f32 = 1.000000
|
| 38 |
+
llama_model_loader: - kv 26: laguna.rope.scaling.yarn_beta_fast f32 = 64.000000
|
| 39 |
+
llama_model_loader: - kv 27: laguna.rope.scaling.yarn_beta_slow f32 = 1.000000
|
| 40 |
+
llama_model_loader: - kv 28: laguna.expert_count u32 = 256
|
| 41 |
+
llama_model_loader: - kv 29: laguna.expert_used_count u32 = 16
|
| 42 |
+
llama_model_loader: - kv 30: laguna.expert_feed_forward_length u32 = 1024
|
| 43 |
+
llama_model_loader: - kv 31: laguna.expert_shared_feed_forward_length u32 = 1024
|
| 44 |
+
llama_model_loader: - kv 32: laguna.expert_weights_scale f32 = 1.000000
|
| 45 |
+
llama_model_loader: - kv 33: laguna.expert_weights_norm bool = true
|
| 46 |
+
llama_model_loader: - kv 34: laguna.expert_gating_func u32 = 2
|
| 47 |
+
llama_model_loader: - kv 35: laguna.leading_dense_block_count u32 = 3
|
| 48 |
+
llama_model_loader: - kv 36: tokenizer.ggml.model str = gpt2
|
| 49 |
+
llama_model_loader: - kv 37: tokenizer.ggml.pre str = laguna
|
| 50 |
+
llama_model_loader: - kv 38: tokenizer.ggml.tokens arr[str,100352] = ["γ|UNK|γ", "γ|CODE_START|γ",...
|
| 51 |
+
llama_model_loader: - kv 39: tokenizer.ggml.token_type arr[i32,100352] = [3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, ...
|
| 52 |
+
llama_model_loader: - kv 40: tokenizer.ggml.merges arr[str,100026] = ["i n", "Δ t", "Δ Δ ", "e r", "Δ a...
|
| 53 |
+
llama_model_loader: - kv 41: tokenizer.ggml.bos_token_id u32 = 2
|
| 54 |
+
llama_model_loader: - kv 42: tokenizer.ggml.eos_token_id u32 = 2
|
| 55 |
+
llama_model_loader: - kv 43: tokenizer.ggml.unknown_token_id u32 = 0
|
| 56 |
+
llama_model_loader: - kv 44: tokenizer.ggml.seperator_token_id u32 = 8
|
| 57 |
+
llama_model_loader: - kv 45: tokenizer.ggml.padding_token_id u32 = 9
|
| 58 |
+
llama_model_loader: - kv 46: tokenizer.ggml.mask_token_id u32 = 12
|
| 59 |
+
llama_model_loader: - kv 47: tokenizer.ggml.add_bos_token bool = true
|
| 60 |
+
llama_model_loader: - kv 48: tokenizer.ggml.add_sep_token bool = true
|
| 61 |
+
llama_model_loader: - kv 49: tokenizer.chat_template str = {#- Copied from laguna_glm_thinking_v...
|
| 62 |
+
llama_model_loader: - kv 50: tokenizer.ggml.eot_token_id u32 = 24
|
| 63 |
+
llama_model_loader: - kv 51: general.quantization_version u32 = 2
|
| 64 |
+
llama_model_loader: - kv 52: quantize.imatrix.file str = /mnt/pool/gguf/laguna-m1/imatrix/q8-p...
|
| 65 |
+
llama_model_loader: - kv 53: quantize.imatrix.dataset str = /workspace/laguna-m1/corpus/laguna-m1...
|
| 66 |
+
llama_model_loader: - kv 54: quantize.imatrix.entries_count i32 = 828
|
| 67 |
+
llama_model_loader: - kv 55: quantize.imatrix.chunks_count i32 = 2048
|
| 68 |
+
llama_model_loader: - kv 56: split.no u16 = 0
|
| 69 |
+
llama_model_loader: - kv 57: split.count u16 = 10
|
| 70 |
+
llama_model_loader: - kv 58: split.tensors.count i32 = 1178
|
| 71 |
+
llama_model_loader: - type f32: 415 tensors
|
| 72 |
+
llama_model_loader: - type q8_0: 140 tensors
|
| 73 |
+
llama_model_loader: - type q5_K: 70 tensors
|
| 74 |
+
llama_model_loader: - type q6_K: 2 tensors
|
| 75 |
+
llama_model_loader: - type iq3_s: 551 tensors
|
| 76 |
+
load: 0 unused tokens
|
| 77 |
+
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 78 |
+
load: special_eot_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 79 |
+
load: printing all EOG tokens:
|
| 80 |
+
load: - 2 ('γ|EOS|γ')
|
| 81 |
+
load: - 24 ('</assistant>')
|
| 82 |
+
load: special tokens cache size = 70
|
| 83 |
+
load: token to piece cache size = 0.6432 MB
|
| 84 |
+
llm_load_print_meta: format = GGUF V3 (latest)
|
| 85 |
+
llm_load_print_meta: arch = laguna
|
| 86 |
+
llm_load_print_meta: n_ctx_train = 262144
|
| 87 |
+
llm_load_print_meta: n_embd = 4096
|
| 88 |
+
llm_load_print_meta: n_layer = 70
|
| 89 |
+
llm_load_print_meta: n_head = 64
|
| 90 |
+
llm_load_print_meta: n_head_kv = 8
|
| 91 |
+
llm_load_print_meta: n_rot = 128
|
| 92 |
+
llm_load_print_meta: n_swa = 0
|
| 93 |
+
llm_load_print_meta: n_swa_pattern = 1
|
| 94 |
+
llm_load_print_meta: n_embd_head_k = 128
|
| 95 |
+
llm_load_print_meta: n_embd_head_v = 128
|
| 96 |
+
llm_load_print_meta: n_gqa = 8
|
| 97 |
+
llm_load_print_meta: n_embd_k_gqa = 1024
|
| 98 |
+
llm_load_print_meta: n_embd_v_gqa = 1024
|
| 99 |
+
llm_load_print_meta: f_norm_eps = 0.0e+00
|
| 100 |
+
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
|
| 101 |
+
llm_load_print_meta: f_clamp_kqv = 0.0e+00
|
| 102 |
+
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
|
| 103 |
+
llm_load_print_meta: f_logit_scale = 0.0e+00
|
| 104 |
+
llm_load_print_meta: n_ff = 16384
|
| 105 |
+
llm_load_print_meta: n_expert = 256
|
| 106 |
+
llm_load_print_meta: n_expert_used = 16
|
| 107 |
+
llm_load_print_meta: causal attn = 1
|
| 108 |
+
llm_load_print_meta: pooling type = 0
|
| 109 |
+
llm_load_print_meta: rope type = 2
|
| 110 |
+
llm_load_print_meta: rope scaling = yarn
|
| 111 |
+
llm_load_print_meta: freq_base_train = 500000.0
|
| 112 |
+
llm_load_print_meta: freq_scale_train = 0.015625
|
| 113 |
+
llm_load_print_meta: n_ctx_orig_yarn = 4096
|
| 114 |
+
llm_load_print_meta: rope_finetuned = unknown
|
| 115 |
+
llm_load_print_meta: ssm_d_conv = 0
|
| 116 |
+
llm_load_print_meta: ssm_d_inner = 0
|
| 117 |
+
llm_load_print_meta: ssm_d_state = 0
|
| 118 |
+
llm_load_print_meta: ssm_dt_rank = 0
|
| 119 |
+
llm_load_print_meta: ssm_n_group = 0
|
| 120 |
+
llm_load_print_meta: model type = ?B
|
| 121 |
+
llm_load_print_meta: model ftype = IQ3_S - 3.4375 bpw
|
| 122 |
+
llm_load_print_meta: model params = 225.796 B
|
| 123 |
+
llm_load_print_meta: model size = 91.803 GiB (3.492 BPW)
|
| 124 |
+
llm_load_print_meta: repeating layers = 91.175 GiB (3.481 BPW, 224.974 B parameters)
|
| 125 |
+
llm_load_print_meta: general.name = Laguna M.1
|
| 126 |
+
print_info: vocab type = BPE
|
| 127 |
+
print_info: n_vocab = 100352
|
| 128 |
+
print_info: n_merges = 100026
|
| 129 |
+
print_info: BOS token = 2 'γ|EOS|γ'
|
| 130 |
+
print_info: EOS token = 2 'γ|EOS|γ'
|
| 131 |
+
print_info: EOT token = 24 '</assistant>'
|
| 132 |
+
print_info: UNK token = 0 'γ|UNK|γ'
|
| 133 |
+
print_info: SEP token = 8 'γ|SEP|γ'
|
| 134 |
+
print_info: PAD token = 9 'γ|PAD|γ'
|
| 135 |
+
print_info: MASK token = 12 'γ|MASK|γ'
|
| 136 |
+
print_info: LF token = 268 'Δ'
|
| 137 |
+
print_info: EOG token = 2 'γ|EOS|γ'
|
| 138 |
+
print_info: EOG token = 24 '</assistant>'
|
| 139 |
+
print_info: max token length = 830
|
| 140 |
+
======================================= HAVE_FANCY_SIMD is NOT defined
|
| 141 |
+
------------------- Layer sizes:
|
| 142 |
+
Layer 0: 140.53, 16.00, 156.53 22.00 MiB
|
| 143 |
+
Layer 1: 140.53, 16.00, 156.53 22.00 MiB
|
| 144 |
+
Layer 2: 140.53, 16.00, 156.53 22.00 MiB
|
| 145 |
+
Layer 3: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 146 |
+
Layer 4: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 147 |
+
Layer 5: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 148 |
+
Layer 6: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 149 |
+
Layer 7: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 150 |
+
Layer 8: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 151 |
+
Layer 9: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 152 |
+
Layer 10: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 153 |
+
Layer 11: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 154 |
+
Layer 12: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 155 |
+
Layer 13: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 156 |
+
Layer 14: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 157 |
+
Layer 15: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 158 |
+
Layer 16: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 159 |
+
Layer 17: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 160 |
+
Layer 18: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 161 |
+
Layer 19: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 162 |
+
Layer 20: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 163 |
+
Layer 21: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 164 |
+
Layer 22: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 165 |
+
Layer 23: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 166 |
+
Layer 24: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 167 |
+
Layer 25: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 168 |
+
Layer 26: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 169 |
+
Layer 27: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 170 |
+
Layer 28: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 171 |
+
Layer 29: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 172 |
+
Layer 30: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 173 |
+
Layer 31: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 174 |
+
Layer 32: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 175 |
+
Layer 33: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 176 |
+
Layer 34: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 177 |
+
Layer 35: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 178 |
+
Layer 36: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 179 |
+
Layer 37: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 180 |
+
Layer 38: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 181 |
+
Layer 39: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 182 |
+
Layer 40: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 183 |
+
Layer 41: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 184 |
+
Layer 42: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 185 |
+
Layer 43: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 186 |
+
Layer 44: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 187 |
+
Layer 45: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 188 |
+
Layer 46: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 189 |
+
Layer 47: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 190 |
+
Layer 48: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 191 |
+
Layer 49: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 192 |
+
Layer 50: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 193 |
+
Layer 51: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 194 |
+
Layer 52: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 195 |
+
Layer 53: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 196 |
+
Layer 54: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 197 |
+
Layer 55: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 198 |
+
Layer 56: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 199 |
+
Layer 57: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 200 |
+
Layer 58: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 201 |
+
Layer 59: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 202 |
+
Layer 60: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 203 |
+
Layer 61: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 204 |
+
Layer 62: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 205 |
+
Layer 63: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 206 |
+
Layer 64: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 207 |
+
Layer 65: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 208 |
+
Layer 66: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 209 |
+
Layer 67: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 210 |
+
Layer 68: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 211 |
+
Layer 69: 1387.19, 16.00, 1403.19 52.00 MiB
|
| 212 |
+
Layer 70: 321.56, 0.00, 321.56 MiB (output layer)
|
| 213 |
+
--------------------------------------------------------------------------
|
| 214 |
+
Total : 93363.29, 1120.00, 94483.29 MiB
|
| 215 |
+
Memory required for model tensors + cache: 94805 MiB
|
| 216 |
+
Memory available on all devices - compute: 95379 MiB
|
| 217 |
+
llm_load_tensors: ggml ctx size = 1.02 MiB
|
| 218 |
+
llm_load_tensors: offloading 70 repeating layers to GPU
|
| 219 |
+
llm_load_tensors: offloading non-repeating layers to GPU
|
| 220 |
+
llm_load_tensors: offloaded 71/71 layers to GPU
|
| 221 |
+
llm_load_tensors: CPU buffer size = 321.56 MiB
|
| 222 |
+
llm_load_tensors: CUDA0 buffer size = 93684.87 MiB
|
| 223 |
+
....................................................................................................
|
| 224 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 225 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 226 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 227 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 228 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 229 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 230 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 231 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 232 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 233 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 234 |
+
llm_load_tensors: dense parameters loaded in 110.62s (5.44 GiB), expert parameters deferred (86.37 GiB)
|
| 235 |
+
llama_init_from_model: n_ctx = 4096
|
| 236 |
+
llama_init_from_model: n_batch = 512
|
| 237 |
+
llama_init_from_model: n_ubatch = 128
|
| 238 |
+
llama_init_from_model: flash_attn = 1
|
| 239 |
+
llama_init_from_model: attn_max_b = 0
|
| 240 |
+
llama_init_from_model: fused_moe = 1
|
| 241 |
+
llama_init_from_model: grouped er = 0
|
| 242 |
+
llama_init_from_model: fused_up_gate = 1
|
| 243 |
+
llama_init_from_model: fused_mmad = 1
|
| 244 |
+
llama_init_from_model: rope_cache = 0
|
| 245 |
+
llama_init_from_model: graph_reuse = 1
|
| 246 |
+
llama_init_from_model: k_cache_hadam = 0
|
| 247 |
+
llama_init_from_model: v_cache_hadam = 0
|
| 248 |
+
llama_init_from_model: split_mode_graph_scheduling = 0
|
| 249 |
+
llama_init_from_model: reduce_type = f16
|
| 250 |
+
llama_init_from_model: sched_async = 0
|
| 251 |
+
llama_init_from_model: ser = -1, 0
|
| 252 |
+
llama_init_from_model: freq_base = 500000.0
|
| 253 |
+
llama_init_from_model: freq_scale = 0.015625
|
| 254 |
+
llama_kv_cache_init: CUDA0 KV buffer size = 1120.00 MiB
|
| 255 |
+
llama_init_from_model: KV self size = 1120.00 MiB, K (f16): 560.00 MiB, V (f16): 560.00 MiB
|
| 256 |
+
llama_init_from_model: CUDA_Host output buffer size = 0.38 MiB
|
| 257 |
+
llama_init_from_model: CUDA0 compute buffer size = 51.00 MiB
|
| 258 |
+
llama_init_from_model: CUDA_Host compute buffer size = 3.00 MiB
|
| 259 |
+
llama_init_from_model: graph nodes = 3174
|
| 260 |
+
llama_init_from_model: graph splits = 2
|
| 261 |
+
llama_init_from_model: enabling only_active_experts scheduling
|
| 262 |
+
|
| 263 |
+
system_info: n_threads = 20 / 20 | AVX = 0 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 0 | NEON = 1 | SVE = 0 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | MATMUL_INT8 = 0 |
|
| 264 |
+
perplexity: tokenizing the input ..
|
| 265 |
+
perplexity: tokenization took 1789.63 ms
|
| 266 |
+
perplexity: calculating perplexity over 8 chunks, n_ctx=4096, batch_size=512, n_seq=1
|
| 267 |
+
perplexity: 25.98 seconds per pass - ETA 3.45 minutes
|
| 268 |
+
[1]1.3713,[2]2.2535,[3]1.8046,[4]2.4539,[5]2.6550,[6]2.6851,[7]2.6250,[8]2.4683,
|
| 269 |
+
llama_print_timings: load time = 128307.95 ms
|
| 270 |
+
llama_print_timings: sample time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
| 271 |
+
llama_print_timings: prompt eval time = 209390.13 ms / 32768 tokens ( 6.39 ms per token, 156.49 tokens per second)
|
| 272 |
+
llama_print_timings: eval time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
| 273 |
+
llama_print_timings: total time = 322931.50 ms / 32769 tokens
|
| 274 |
+
~ggml_backend_cuda_context: have 1 graphs
|
| 275 |
+
|
| 276 |
+
Final estimate: PPL over 8 chunks for n_ctx=4096 = 2.4683 +/- 0.04051
|
IQ3_S/quantize-command.txt
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
quant=IQ3_S
|
| 2 |
+
lane=vanilla-core
|
| 3 |
+
started_at=2026-06-21T01:22:11-04:00
|
| 4 |
+
bf16_first=/home/janet/laguna-gguf/laguna-m1/bf16/Laguna-M.1-BF16-00001-of-00010.gguf
|
| 5 |
+
imatrix=/mnt/pool/gguf/laguna-m1/imatrix/q8-public-v1-256-20260620-231949/laguna-m1-q8-public-v1-256-gatefix.imatrix
|
| 6 |
+
corpus=/mnt/pool/gguf/laguna-m1/calib/laguna-m1-imatrix-calibration-v1.txt
|
| 7 |
+
threads=20
|
| 8 |
+
command=/home/janet/src/ik_llama.cpp/build-cuda/bin/llama-quantize --keep-split --partial-requant --imatrix /mnt/pool/gguf/laguna-m1/imatrix/q8-public-v1-256-20260620-231949/laguna-m1-q8-public-v1-256-gatefix.imatrix --output-tensor-type q6_K --token-embedding-type q6_K /home/janet/laguna-gguf/laguna-m1/bf16/Laguna-M.1-BF16-00001-of-00010.gguf /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1.gguf IQ3_S 20
|
IQ3_S/quantize.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
IQ3_S/result.tsv
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
quant lane status files bytes sha256sums ppl ppl_unc kld kld_unc kl_reference completed_at log
|
| 2 |
+
IQ3_S vanilla-core finished 10 98576647840 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/SHA256SUMS 2.4683 0.04051 pending pending pending 2026-06-21T02:40:21-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/quantize.log
|
IQ3_S/smoke.log
ADDED
|
@@ -0,0 +1,285 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Log start
|
| 2 |
+
main: build = 4641 (be7d53ce)
|
| 3 |
+
main: built with cc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0 for aarch64-linux-gnu
|
| 4 |
+
main: seed = 123
|
| 5 |
+
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
|
| 6 |
+
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
|
| 7 |
+
ggml_cuda_init: found 1 CUDA devices:
|
| 8 |
+
Device 0: NVIDIA GB10, compute capability 12.1, VMM: yes, VRAM: 124610 MiB
|
| 9 |
+
CUDA0: using device CUDA0 - 12326 MiB free
|
| 10 |
+
llama_model_loader: additional 9 GGUFs metadata loaded.
|
| 11 |
+
llama_model_loader: loaded meta data with 59 key-value pairs and 1178 tensors from /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf (version GGUF V3 (latest))
|
| 12 |
+
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
|
| 13 |
+
llama_model_loader: - kv 0: general.architecture str = laguna
|
| 14 |
+
llama_model_loader: - kv 1: general.type str = model
|
| 15 |
+
llama_model_loader: - kv 2: general.name str = Laguna M.1
|
| 16 |
+
llama_model_loader: - kv 3: general.size_label str = 256x11B
|
| 17 |
+
llama_model_loader: - kv 4: general.license str = apache-2.0
|
| 18 |
+
llama_model_loader: - kv 5: general.tags arr[str,6] = ["laguna-m.1", "vllm", "sglang", "bf1...
|
| 19 |
+
llama_model_loader: - kv 6: laguna.context_length u32 = 262144
|
| 20 |
+
llama_model_loader: - kv 7: laguna.embedding_length u32 = 4096
|
| 21 |
+
llama_model_loader: - kv 8: laguna.block_count u32 = 70
|
| 22 |
+
llama_model_loader: - kv 9: laguna.feed_forward_length u32 = 16384
|
| 23 |
+
llama_model_loader: - kv 10: laguna.attention.head_count arr[i32,70] = [64, 64, 64, 64, 64, 64, 64, 64, 64, ...
|
| 24 |
+
llama_model_loader: - kv 11: laguna.attention.head_count_kv u32 = 8
|
| 25 |
+
llama_model_loader: - kv 12: laguna.attention.key_length u32 = 128
|
| 26 |
+
llama_model_loader: - kv 13: laguna.attention.value_length u32 = 128
|
| 27 |
+
llama_model_loader: - kv 14: laguna.attention.layer_norm_rms_epsilon f32 = 0.000001
|
| 28 |
+
llama_model_loader: - kv 15: general.file_type u32 = 26
|
| 29 |
+
llama_model_loader: - kv 16: laguna.attention.sliding_window u32 = 0
|
| 30 |
+
llama_model_loader: - kv 17: laguna.rope.dimension_count u32 = 128
|
| 31 |
+
llama_model_loader: - kv 18: laguna.rope.dimension_count_swa u32 = 128
|
| 32 |
+
llama_model_loader: - kv 19: laguna.rope.freq_base f32 = 500000.000000
|
| 33 |
+
llama_model_loader: - kv 20: laguna.rope.freq_base_swa f32 = 10000.000000
|
| 34 |
+
llama_model_loader: - kv 21: laguna.rope.scaling.type str = yarn
|
| 35 |
+
llama_model_loader: - kv 22: laguna.rope.scaling.factor f32 = 64.000000
|
| 36 |
+
llama_model_loader: - kv 23: laguna.rope.scaling.original_context_length u32 = 4096
|
| 37 |
+
llama_model_loader: - kv 24: laguna.rope.scaling.yarn_ext_factor f32 = 1.000000
|
| 38 |
+
llama_model_loader: - kv 25: laguna.rope.scaling.yarn_attn_factor f32 = 1.000000
|
| 39 |
+
llama_model_loader: - kv 26: laguna.rope.scaling.yarn_beta_fast f32 = 64.000000
|
| 40 |
+
llama_model_loader: - kv 27: laguna.rope.scaling.yarn_beta_slow f32 = 1.000000
|
| 41 |
+
llama_model_loader: - kv 28: laguna.expert_count u32 = 256
|
| 42 |
+
llama_model_loader: - kv 29: laguna.expert_used_count u32 = 16
|
| 43 |
+
llama_model_loader: - kv 30: laguna.expert_feed_forward_length u32 = 1024
|
| 44 |
+
llama_model_loader: - kv 31: laguna.expert_shared_feed_forward_length u32 = 1024
|
| 45 |
+
llama_model_loader: - kv 32: laguna.expert_weights_scale f32 = 1.000000
|
| 46 |
+
llama_model_loader: - kv 33: laguna.expert_weights_norm bool = true
|
| 47 |
+
llama_model_loader: - kv 34: laguna.expert_gating_func u32 = 2
|
| 48 |
+
llama_model_loader: - kv 35: laguna.leading_dense_block_count u32 = 3
|
| 49 |
+
llama_model_loader: - kv 36: tokenizer.ggml.model str = gpt2
|
| 50 |
+
llama_model_loader: - kv 37: tokenizer.ggml.pre str = laguna
|
| 51 |
+
llama_model_loader: - kv 38: tokenizer.ggml.tokens arr[str,100352] = ["γ|UNK|γ", "γ|CODE_START|γ",...
|
| 52 |
+
llama_model_loader: - kv 39: tokenizer.ggml.token_type arr[i32,100352] = [3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, ...
|
| 53 |
+
llama_model_loader: - kv 40: tokenizer.ggml.merges arr[str,100026] = ["i n", "Δ t", "Δ Δ ", "e r", "Δ a...
|
| 54 |
+
llama_model_loader: - kv 41: tokenizer.ggml.bos_token_id u32 = 2
|
| 55 |
+
llama_model_loader: - kv 42: tokenizer.ggml.eos_token_id u32 = 2
|
| 56 |
+
llama_model_loader: - kv 43: tokenizer.ggml.unknown_token_id u32 = 0
|
| 57 |
+
llama_model_loader: - kv 44: tokenizer.ggml.seperator_token_id u32 = 8
|
| 58 |
+
llama_model_loader: - kv 45: tokenizer.ggml.padding_token_id u32 = 9
|
| 59 |
+
llama_model_loader: - kv 46: tokenizer.ggml.mask_token_id u32 = 12
|
| 60 |
+
llama_model_loader: - kv 47: tokenizer.ggml.add_bos_token bool = true
|
| 61 |
+
llama_model_loader: - kv 48: tokenizer.ggml.add_sep_token bool = true
|
| 62 |
+
llama_model_loader: - kv 49: tokenizer.chat_template str = {#- Copied from laguna_glm_thinking_v...
|
| 63 |
+
llama_model_loader: - kv 50: tokenizer.ggml.eot_token_id u32 = 24
|
| 64 |
+
llama_model_loader: - kv 51: general.quantization_version u32 = 2
|
| 65 |
+
llama_model_loader: - kv 52: quantize.imatrix.file str = /mnt/pool/gguf/laguna-m1/imatrix/q8-p...
|
| 66 |
+
llama_model_loader: - kv 53: quantize.imatrix.dataset str = /workspace/laguna-m1/corpus/laguna-m1...
|
| 67 |
+
llama_model_loader: - kv 54: quantize.imatrix.entries_count i32 = 828
|
| 68 |
+
llama_model_loader: - kv 55: quantize.imatrix.chunks_count i32 = 2048
|
| 69 |
+
llama_model_loader: - kv 56: split.no u16 = 0
|
| 70 |
+
llama_model_loader: - kv 57: split.count u16 = 10
|
| 71 |
+
llama_model_loader: - kv 58: split.tensors.count i32 = 1178
|
| 72 |
+
llama_model_loader: - type f32: 415 tensors
|
| 73 |
+
llama_model_loader: - type q8_0: 140 tensors
|
| 74 |
+
llama_model_loader: - type q5_K: 70 tensors
|
| 75 |
+
llama_model_loader: - type q6_K: 2 tensors
|
| 76 |
+
llama_model_loader: - type iq3_s: 551 tensors
|
| 77 |
+
load: 0 unused tokens
|
| 78 |
+
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 79 |
+
load: special_eot_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 80 |
+
load: printing all EOG tokens:
|
| 81 |
+
load: - 2 ('γ|EOS|γ')
|
| 82 |
+
load: - 24 ('</assistant>')
|
| 83 |
+
load: special tokens cache size = 70
|
| 84 |
+
load: token to piece cache size = 0.6432 MB
|
| 85 |
+
llm_load_print_meta: format = GGUF V3 (latest)
|
| 86 |
+
llm_load_print_meta: arch = laguna
|
| 87 |
+
llm_load_print_meta: n_ctx_train = 262144
|
| 88 |
+
llm_load_print_meta: n_embd = 4096
|
| 89 |
+
llm_load_print_meta: n_layer = 70
|
| 90 |
+
llm_load_print_meta: n_head = 64
|
| 91 |
+
llm_load_print_meta: n_head_kv = 8
|
| 92 |
+
llm_load_print_meta: n_rot = 128
|
| 93 |
+
llm_load_print_meta: n_swa = 0
|
| 94 |
+
llm_load_print_meta: n_swa_pattern = 1
|
| 95 |
+
llm_load_print_meta: n_embd_head_k = 128
|
| 96 |
+
llm_load_print_meta: n_embd_head_v = 128
|
| 97 |
+
llm_load_print_meta: n_gqa = 8
|
| 98 |
+
llm_load_print_meta: n_embd_k_gqa = 1024
|
| 99 |
+
llm_load_print_meta: n_embd_v_gqa = 1024
|
| 100 |
+
llm_load_print_meta: f_norm_eps = 0.0e+00
|
| 101 |
+
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
|
| 102 |
+
llm_load_print_meta: f_clamp_kqv = 0.0e+00
|
| 103 |
+
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
|
| 104 |
+
llm_load_print_meta: f_logit_scale = 0.0e+00
|
| 105 |
+
llm_load_print_meta: n_ff = 16384
|
| 106 |
+
llm_load_print_meta: n_expert = 256
|
| 107 |
+
llm_load_print_meta: n_expert_used = 16
|
| 108 |
+
llm_load_print_meta: causal attn = 1
|
| 109 |
+
llm_load_print_meta: pooling type = 0
|
| 110 |
+
llm_load_print_meta: rope type = 2
|
| 111 |
+
llm_load_print_meta: rope scaling = yarn
|
| 112 |
+
llm_load_print_meta: freq_base_train = 500000.0
|
| 113 |
+
llm_load_print_meta: freq_scale_train = 0.015625
|
| 114 |
+
llm_load_print_meta: n_ctx_orig_yarn = 4096
|
| 115 |
+
llm_load_print_meta: rope_finetuned = unknown
|
| 116 |
+
llm_load_print_meta: ssm_d_conv = 0
|
| 117 |
+
llm_load_print_meta: ssm_d_inner = 0
|
| 118 |
+
llm_load_print_meta: ssm_d_state = 0
|
| 119 |
+
llm_load_print_meta: ssm_dt_rank = 0
|
| 120 |
+
llm_load_print_meta: ssm_n_group = 0
|
| 121 |
+
llm_load_print_meta: model type = ?B
|
| 122 |
+
llm_load_print_meta: model ftype = IQ3_S - 3.4375 bpw
|
| 123 |
+
llm_load_print_meta: model params = 225.796 B
|
| 124 |
+
llm_load_print_meta: model size = 91.803 GiB (3.492 BPW)
|
| 125 |
+
llm_load_print_meta: repeating layers = 91.175 GiB (3.481 BPW, 224.974 B parameters)
|
| 126 |
+
llm_load_print_meta: general.name = Laguna M.1
|
| 127 |
+
print_info: vocab type = BPE
|
| 128 |
+
print_info: n_vocab = 100352
|
| 129 |
+
print_info: n_merges = 100026
|
| 130 |
+
print_info: BOS token = 2 'γ|EOS|γ'
|
| 131 |
+
print_info: EOS token = 2 'γ|EOS|γ'
|
| 132 |
+
print_info: EOT token = 24 '</assistant>'
|
| 133 |
+
print_info: UNK token = 0 'γ|UNK|γ'
|
| 134 |
+
print_info: SEP token = 8 'γ|SEP|γ'
|
| 135 |
+
print_info: PAD token = 9 'γ|PAD|γ'
|
| 136 |
+
print_info: MASK token = 12 'γ|MASK|γ'
|
| 137 |
+
print_info: LF token = 268 'Δ'
|
| 138 |
+
print_info: EOG token = 2 'γ|EOS|γ'
|
| 139 |
+
print_info: EOG token = 24 '</assistant>'
|
| 140 |
+
print_info: max token length = 830
|
| 141 |
+
======================================= HAVE_FANCY_SIMD is NOT defined
|
| 142 |
+
------------------- Layer sizes:
|
| 143 |
+
Layer 0: 140.53, 8.00, 148.53 22.00 MiB
|
| 144 |
+
Layer 1: 140.53, 8.00, 148.53 22.00 MiB
|
| 145 |
+
Layer 2: 140.53, 8.00, 148.53 22.00 MiB
|
| 146 |
+
Layer 3: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 147 |
+
Layer 4: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 148 |
+
Layer 5: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 149 |
+
Layer 6: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 150 |
+
Layer 7: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 151 |
+
Layer 8: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 152 |
+
Layer 9: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 153 |
+
Layer 10: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 154 |
+
Layer 11: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 155 |
+
Layer 12: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 156 |
+
Layer 13: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 157 |
+
Layer 14: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 158 |
+
Layer 15: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 159 |
+
Layer 16: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 160 |
+
Layer 17: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 161 |
+
Layer 18: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 162 |
+
Layer 19: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 163 |
+
Layer 20: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 164 |
+
Layer 21: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 165 |
+
Layer 22: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 166 |
+
Layer 23: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 167 |
+
Layer 24: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 168 |
+
Layer 25: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 169 |
+
Layer 26: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 170 |
+
Layer 27: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 171 |
+
Layer 28: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 172 |
+
Layer 29: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 173 |
+
Layer 30: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 174 |
+
Layer 31: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 175 |
+
Layer 32: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 176 |
+
Layer 33: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 177 |
+
Layer 34: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 178 |
+
Layer 35: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 179 |
+
Layer 36: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 180 |
+
Layer 37: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 181 |
+
Layer 38: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 182 |
+
Layer 39: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 183 |
+
Layer 40: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 184 |
+
Layer 41: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 185 |
+
Layer 42: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 186 |
+
Layer 43: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 187 |
+
Layer 44: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 188 |
+
Layer 45: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 189 |
+
Layer 46: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 190 |
+
Layer 47: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 191 |
+
Layer 48: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 192 |
+
Layer 49: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 193 |
+
Layer 50: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 194 |
+
Layer 51: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 195 |
+
Layer 52: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 196 |
+
Layer 53: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 197 |
+
Layer 54: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 198 |
+
Layer 55: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 199 |
+
Layer 56: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 200 |
+
Layer 57: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 201 |
+
Layer 58: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 202 |
+
Layer 59: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 203 |
+
Layer 60: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 204 |
+
Layer 61: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 205 |
+
Layer 62: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 206 |
+
Layer 63: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 207 |
+
Layer 64: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 208 |
+
Layer 65: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 209 |
+
Layer 66: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 210 |
+
Layer 67: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 211 |
+
Layer 68: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 212 |
+
Layer 69: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 213 |
+
Layer 70: 321.56, 0.00, 321.56 MiB (output layer)
|
| 214 |
+
--------------------------------------------------------------------------
|
| 215 |
+
Total : 93363.29, 560.00, 93923.29 MiB
|
| 216 |
+
Memory required for model tensors + cache: 94245 MiB
|
| 217 |
+
Memory available on all devices - compute: 11206 MiB
|
| 218 |
+
llm_load_tensors: ggml ctx size = 1.02 MiB
|
| 219 |
+
llm_load_tensors: offloading 70 repeating layers to GPU
|
| 220 |
+
llm_load_tensors: offloading non-repeating layers to GPU
|
| 221 |
+
llm_load_tensors: offloaded 71/71 layers to GPU
|
| 222 |
+
llm_load_tensors: CPU buffer size = 321.56 MiB
|
| 223 |
+
llm_load_tensors: CUDA0 buffer size = 93684.87 MiB
|
| 224 |
+
....................................................................................................
|
| 225 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 226 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 227 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 228 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 229 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 230 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 231 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 232 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 233 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 234 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 235 |
+
llm_load_tensors: dense parameters loaded in 125.48s (5.44 GiB), expert parameters deferred (86.37 GiB)
|
| 236 |
+
llama_init_from_model: n_ctx = 2048
|
| 237 |
+
llama_init_from_model: n_batch = 256
|
| 238 |
+
llama_init_from_model: n_ubatch = 128
|
| 239 |
+
llama_init_from_model: flash_attn = 1
|
| 240 |
+
llama_init_from_model: attn_max_b = 0
|
| 241 |
+
llama_init_from_model: fused_moe = 1
|
| 242 |
+
llama_init_from_model: grouped er = 0
|
| 243 |
+
llama_init_from_model: fused_up_gate = 1
|
| 244 |
+
llama_init_from_model: fused_mmad = 1
|
| 245 |
+
llama_init_from_model: rope_cache = 0
|
| 246 |
+
llama_init_from_model: graph_reuse = 1
|
| 247 |
+
llama_init_from_model: k_cache_hadam = 0
|
| 248 |
+
llama_init_from_model: v_cache_hadam = 0
|
| 249 |
+
llama_init_from_model: split_mode_graph_scheduling = 0
|
| 250 |
+
llama_init_from_model: reduce_type = f16
|
| 251 |
+
llama_init_from_model: sched_async = 0
|
| 252 |
+
llama_init_from_model: ser = -1, 0
|
| 253 |
+
llama_init_from_model: freq_base = 500000.0
|
| 254 |
+
llama_init_from_model: freq_scale = 0.015625
|
| 255 |
+
llama_kv_cache_init: CUDA0 KV buffer size = 560.00 MiB
|
| 256 |
+
llama_init_from_model: KV self size = 560.00 MiB, K (f16): 280.00 MiB, V (f16): 280.00 MiB
|
| 257 |
+
llama_init_from_model: CUDA_Host output buffer size = 0.38 MiB
|
| 258 |
+
llama_init_from_model: CUDA0 compute buffer size = 51.00 MiB
|
| 259 |
+
llama_init_from_model: CUDA_Host compute buffer size = 2.50 MiB
|
| 260 |
+
llama_init_from_model: graph nodes = 3174
|
| 261 |
+
llama_init_from_model: graph splits = 2
|
| 262 |
+
llama_init_from_model: enabling only_active_experts scheduling
|
| 263 |
+
|
| 264 |
+
system_info: n_threads = 20 / 20 | AVX = 0 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 0 | NEON = 1 | SVE = 0 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | MATMUL_INT8 = 0 |
|
| 265 |
+
sampling:
|
| 266 |
+
repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
|
| 267 |
+
top_k = 40, tfs_z = 1.000, top_p = 0.950, min_p = 0.050, typical_p = 1.000, temp = 0.200
|
| 268 |
+
mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
|
| 269 |
+
xtc_probability = 0.000, xtc_threshold = 1.000, top_n_sigma = 0.000
|
| 270 |
+
adaptive_target = -1.00, adaptive_decay = 0.90
|
| 271 |
+
sampling order:
|
| 272 |
+
CFG -> Penalties -> dry -> top_k -> tfs_z -> typical_p -> top_p -> min_p -> xtc -> top_n_sigma -> temperature -> adaptive_p
|
| 273 |
+
generate: n_ctx = 2048, n_batch = 256, n_predict = 48, n_keep = 1
|
| 274 |
+
|
| 275 |
+
|
| 276 |
+
Write a tiny Python function that computes Fibonacci numbers iteratively. Your function should be named 'fibonacci' and take a single integer argument 'n'. It should return the nth Fibonacci number. The function should handle edge cases where n is 0 or 1.
|
| 277 |
+
|
| 278 |
+
We are computing the Fibonacci sequence iteratively
|
| 279 |
+
llama_print_timings: load time = 126743.48 ms
|
| 280 |
+
llama_print_timings: sample time = 1.94 ms / 48 runs ( 0.04 ms per token, 24742.27 tokens per second)
|
| 281 |
+
llama_print_timings: prompt eval time = 952.45 ms / 12 tokens ( 79.37 ms per token, 12.60 tokens per second)
|
| 282 |
+
llama_print_timings: eval time = 3427.81 ms / 47 runs ( 72.93 ms per token, 13.71 tokens per second)
|
| 283 |
+
llama_print_timings: total time = 130246.48 ms / 59 tokens
|
| 284 |
+
~ggml_backend_cuda_context: have 2 graphs
|
| 285 |
+
Log end
|
IQ3_S/smoke.ngl99.log
ADDED
|
@@ -0,0 +1,285 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Log start
|
| 2 |
+
main: build = 4641 (be7d53ce)
|
| 3 |
+
main: built with cc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0 for aarch64-linux-gnu
|
| 4 |
+
main: seed = 123
|
| 5 |
+
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
|
| 6 |
+
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
|
| 7 |
+
ggml_cuda_init: found 1 CUDA devices:
|
| 8 |
+
Device 0: NVIDIA GB10, compute capability 12.1, VMM: yes, VRAM: 124610 MiB
|
| 9 |
+
CUDA0: using device CUDA0 - 12326 MiB free
|
| 10 |
+
llama_model_loader: additional 9 GGUFs metadata loaded.
|
| 11 |
+
llama_model_loader: loaded meta data with 59 key-value pairs and 1178 tensors from /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf (version GGUF V3 (latest))
|
| 12 |
+
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
|
| 13 |
+
llama_model_loader: - kv 0: general.architecture str = laguna
|
| 14 |
+
llama_model_loader: - kv 1: general.type str = model
|
| 15 |
+
llama_model_loader: - kv 2: general.name str = Laguna M.1
|
| 16 |
+
llama_model_loader: - kv 3: general.size_label str = 256x11B
|
| 17 |
+
llama_model_loader: - kv 4: general.license str = apache-2.0
|
| 18 |
+
llama_model_loader: - kv 5: general.tags arr[str,6] = ["laguna-m.1", "vllm", "sglang", "bf1...
|
| 19 |
+
llama_model_loader: - kv 6: laguna.context_length u32 = 262144
|
| 20 |
+
llama_model_loader: - kv 7: laguna.embedding_length u32 = 4096
|
| 21 |
+
llama_model_loader: - kv 8: laguna.block_count u32 = 70
|
| 22 |
+
llama_model_loader: - kv 9: laguna.feed_forward_length u32 = 16384
|
| 23 |
+
llama_model_loader: - kv 10: laguna.attention.head_count arr[i32,70] = [64, 64, 64, 64, 64, 64, 64, 64, 64, ...
|
| 24 |
+
llama_model_loader: - kv 11: laguna.attention.head_count_kv u32 = 8
|
| 25 |
+
llama_model_loader: - kv 12: laguna.attention.key_length u32 = 128
|
| 26 |
+
llama_model_loader: - kv 13: laguna.attention.value_length u32 = 128
|
| 27 |
+
llama_model_loader: - kv 14: laguna.attention.layer_norm_rms_epsilon f32 = 0.000001
|
| 28 |
+
llama_model_loader: - kv 15: general.file_type u32 = 26
|
| 29 |
+
llama_model_loader: - kv 16: laguna.attention.sliding_window u32 = 0
|
| 30 |
+
llama_model_loader: - kv 17: laguna.rope.dimension_count u32 = 128
|
| 31 |
+
llama_model_loader: - kv 18: laguna.rope.dimension_count_swa u32 = 128
|
| 32 |
+
llama_model_loader: - kv 19: laguna.rope.freq_base f32 = 500000.000000
|
| 33 |
+
llama_model_loader: - kv 20: laguna.rope.freq_base_swa f32 = 10000.000000
|
| 34 |
+
llama_model_loader: - kv 21: laguna.rope.scaling.type str = yarn
|
| 35 |
+
llama_model_loader: - kv 22: laguna.rope.scaling.factor f32 = 64.000000
|
| 36 |
+
llama_model_loader: - kv 23: laguna.rope.scaling.original_context_length u32 = 4096
|
| 37 |
+
llama_model_loader: - kv 24: laguna.rope.scaling.yarn_ext_factor f32 = 1.000000
|
| 38 |
+
llama_model_loader: - kv 25: laguna.rope.scaling.yarn_attn_factor f32 = 1.000000
|
| 39 |
+
llama_model_loader: - kv 26: laguna.rope.scaling.yarn_beta_fast f32 = 64.000000
|
| 40 |
+
llama_model_loader: - kv 27: laguna.rope.scaling.yarn_beta_slow f32 = 1.000000
|
| 41 |
+
llama_model_loader: - kv 28: laguna.expert_count u32 = 256
|
| 42 |
+
llama_model_loader: - kv 29: laguna.expert_used_count u32 = 16
|
| 43 |
+
llama_model_loader: - kv 30: laguna.expert_feed_forward_length u32 = 1024
|
| 44 |
+
llama_model_loader: - kv 31: laguna.expert_shared_feed_forward_length u32 = 1024
|
| 45 |
+
llama_model_loader: - kv 32: laguna.expert_weights_scale f32 = 1.000000
|
| 46 |
+
llama_model_loader: - kv 33: laguna.expert_weights_norm bool = true
|
| 47 |
+
llama_model_loader: - kv 34: laguna.expert_gating_func u32 = 2
|
| 48 |
+
llama_model_loader: - kv 35: laguna.leading_dense_block_count u32 = 3
|
| 49 |
+
llama_model_loader: - kv 36: tokenizer.ggml.model str = gpt2
|
| 50 |
+
llama_model_loader: - kv 37: tokenizer.ggml.pre str = laguna
|
| 51 |
+
llama_model_loader: - kv 38: tokenizer.ggml.tokens arr[str,100352] = ["γ|UNK|γ", "γ|CODE_START|γ",...
|
| 52 |
+
llama_model_loader: - kv 39: tokenizer.ggml.token_type arr[i32,100352] = [3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, ...
|
| 53 |
+
llama_model_loader: - kv 40: tokenizer.ggml.merges arr[str,100026] = ["i n", "Δ t", "Δ Δ ", "e r", "Δ a...
|
| 54 |
+
llama_model_loader: - kv 41: tokenizer.ggml.bos_token_id u32 = 2
|
| 55 |
+
llama_model_loader: - kv 42: tokenizer.ggml.eos_token_id u32 = 2
|
| 56 |
+
llama_model_loader: - kv 43: tokenizer.ggml.unknown_token_id u32 = 0
|
| 57 |
+
llama_model_loader: - kv 44: tokenizer.ggml.seperator_token_id u32 = 8
|
| 58 |
+
llama_model_loader: - kv 45: tokenizer.ggml.padding_token_id u32 = 9
|
| 59 |
+
llama_model_loader: - kv 46: tokenizer.ggml.mask_token_id u32 = 12
|
| 60 |
+
llama_model_loader: - kv 47: tokenizer.ggml.add_bos_token bool = true
|
| 61 |
+
llama_model_loader: - kv 48: tokenizer.ggml.add_sep_token bool = true
|
| 62 |
+
llama_model_loader: - kv 49: tokenizer.chat_template str = {#- Copied from laguna_glm_thinking_v...
|
| 63 |
+
llama_model_loader: - kv 50: tokenizer.ggml.eot_token_id u32 = 24
|
| 64 |
+
llama_model_loader: - kv 51: general.quantization_version u32 = 2
|
| 65 |
+
llama_model_loader: - kv 52: quantize.imatrix.file str = /mnt/pool/gguf/laguna-m1/imatrix/q8-p...
|
| 66 |
+
llama_model_loader: - kv 53: quantize.imatrix.dataset str = /workspace/laguna-m1/corpus/laguna-m1...
|
| 67 |
+
llama_model_loader: - kv 54: quantize.imatrix.entries_count i32 = 828
|
| 68 |
+
llama_model_loader: - kv 55: quantize.imatrix.chunks_count i32 = 2048
|
| 69 |
+
llama_model_loader: - kv 56: split.no u16 = 0
|
| 70 |
+
llama_model_loader: - kv 57: split.count u16 = 10
|
| 71 |
+
llama_model_loader: - kv 58: split.tensors.count i32 = 1178
|
| 72 |
+
llama_model_loader: - type f32: 415 tensors
|
| 73 |
+
llama_model_loader: - type q8_0: 140 tensors
|
| 74 |
+
llama_model_loader: - type q5_K: 70 tensors
|
| 75 |
+
llama_model_loader: - type q6_K: 2 tensors
|
| 76 |
+
llama_model_loader: - type iq3_s: 551 tensors
|
| 77 |
+
load: 0 unused tokens
|
| 78 |
+
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 79 |
+
load: special_eot_id is not in special_eog_ids - the tokenizer config may be incorrect
|
| 80 |
+
load: printing all EOG tokens:
|
| 81 |
+
load: - 2 ('γ|EOS|γ')
|
| 82 |
+
load: - 24 ('</assistant>')
|
| 83 |
+
load: special tokens cache size = 70
|
| 84 |
+
load: token to piece cache size = 0.6432 MB
|
| 85 |
+
llm_load_print_meta: format = GGUF V3 (latest)
|
| 86 |
+
llm_load_print_meta: arch = laguna
|
| 87 |
+
llm_load_print_meta: n_ctx_train = 262144
|
| 88 |
+
llm_load_print_meta: n_embd = 4096
|
| 89 |
+
llm_load_print_meta: n_layer = 70
|
| 90 |
+
llm_load_print_meta: n_head = 64
|
| 91 |
+
llm_load_print_meta: n_head_kv = 8
|
| 92 |
+
llm_load_print_meta: n_rot = 128
|
| 93 |
+
llm_load_print_meta: n_swa = 0
|
| 94 |
+
llm_load_print_meta: n_swa_pattern = 1
|
| 95 |
+
llm_load_print_meta: n_embd_head_k = 128
|
| 96 |
+
llm_load_print_meta: n_embd_head_v = 128
|
| 97 |
+
llm_load_print_meta: n_gqa = 8
|
| 98 |
+
llm_load_print_meta: n_embd_k_gqa = 1024
|
| 99 |
+
llm_load_print_meta: n_embd_v_gqa = 1024
|
| 100 |
+
llm_load_print_meta: f_norm_eps = 0.0e+00
|
| 101 |
+
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
|
| 102 |
+
llm_load_print_meta: f_clamp_kqv = 0.0e+00
|
| 103 |
+
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
|
| 104 |
+
llm_load_print_meta: f_logit_scale = 0.0e+00
|
| 105 |
+
llm_load_print_meta: n_ff = 16384
|
| 106 |
+
llm_load_print_meta: n_expert = 256
|
| 107 |
+
llm_load_print_meta: n_expert_used = 16
|
| 108 |
+
llm_load_print_meta: causal attn = 1
|
| 109 |
+
llm_load_print_meta: pooling type = 0
|
| 110 |
+
llm_load_print_meta: rope type = 2
|
| 111 |
+
llm_load_print_meta: rope scaling = yarn
|
| 112 |
+
llm_load_print_meta: freq_base_train = 500000.0
|
| 113 |
+
llm_load_print_meta: freq_scale_train = 0.015625
|
| 114 |
+
llm_load_print_meta: n_ctx_orig_yarn = 4096
|
| 115 |
+
llm_load_print_meta: rope_finetuned = unknown
|
| 116 |
+
llm_load_print_meta: ssm_d_conv = 0
|
| 117 |
+
llm_load_print_meta: ssm_d_inner = 0
|
| 118 |
+
llm_load_print_meta: ssm_d_state = 0
|
| 119 |
+
llm_load_print_meta: ssm_dt_rank = 0
|
| 120 |
+
llm_load_print_meta: ssm_n_group = 0
|
| 121 |
+
llm_load_print_meta: model type = ?B
|
| 122 |
+
llm_load_print_meta: model ftype = IQ3_S - 3.4375 bpw
|
| 123 |
+
llm_load_print_meta: model params = 225.796 B
|
| 124 |
+
llm_load_print_meta: model size = 91.803 GiB (3.492 BPW)
|
| 125 |
+
llm_load_print_meta: repeating layers = 91.175 GiB (3.481 BPW, 224.974 B parameters)
|
| 126 |
+
llm_load_print_meta: general.name = Laguna M.1
|
| 127 |
+
print_info: vocab type = BPE
|
| 128 |
+
print_info: n_vocab = 100352
|
| 129 |
+
print_info: n_merges = 100026
|
| 130 |
+
print_info: BOS token = 2 'γ|EOS|γ'
|
| 131 |
+
print_info: EOS token = 2 'γ|EOS|γ'
|
| 132 |
+
print_info: EOT token = 24 '</assistant>'
|
| 133 |
+
print_info: UNK token = 0 'γ|UNK|γ'
|
| 134 |
+
print_info: SEP token = 8 'γ|SEP|γ'
|
| 135 |
+
print_info: PAD token = 9 'γ|PAD|γ'
|
| 136 |
+
print_info: MASK token = 12 'γ|MASK|γ'
|
| 137 |
+
print_info: LF token = 268 'Δ'
|
| 138 |
+
print_info: EOG token = 2 'γ|EOS|γ'
|
| 139 |
+
print_info: EOG token = 24 '</assistant>'
|
| 140 |
+
print_info: max token length = 830
|
| 141 |
+
======================================= HAVE_FANCY_SIMD is NOT defined
|
| 142 |
+
------------------- Layer sizes:
|
| 143 |
+
Layer 0: 140.53, 8.00, 148.53 22.00 MiB
|
| 144 |
+
Layer 1: 140.53, 8.00, 148.53 22.00 MiB
|
| 145 |
+
Layer 2: 140.53, 8.00, 148.53 22.00 MiB
|
| 146 |
+
Layer 3: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 147 |
+
Layer 4: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 148 |
+
Layer 5: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 149 |
+
Layer 6: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 150 |
+
Layer 7: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 151 |
+
Layer 8: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 152 |
+
Layer 9: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 153 |
+
Layer 10: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 154 |
+
Layer 11: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 155 |
+
Layer 12: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 156 |
+
Layer 13: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 157 |
+
Layer 14: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 158 |
+
Layer 15: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 159 |
+
Layer 16: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 160 |
+
Layer 17: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 161 |
+
Layer 18: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 162 |
+
Layer 19: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 163 |
+
Layer 20: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 164 |
+
Layer 21: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 165 |
+
Layer 22: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 166 |
+
Layer 23: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 167 |
+
Layer 24: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 168 |
+
Layer 25: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 169 |
+
Layer 26: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 170 |
+
Layer 27: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 171 |
+
Layer 28: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 172 |
+
Layer 29: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 173 |
+
Layer 30: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 174 |
+
Layer 31: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 175 |
+
Layer 32: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 176 |
+
Layer 33: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 177 |
+
Layer 34: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 178 |
+
Layer 35: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 179 |
+
Layer 36: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 180 |
+
Layer 37: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 181 |
+
Layer 38: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 182 |
+
Layer 39: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 183 |
+
Layer 40: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 184 |
+
Layer 41: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 185 |
+
Layer 42: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 186 |
+
Layer 43: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 187 |
+
Layer 44: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 188 |
+
Layer 45: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 189 |
+
Layer 46: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 190 |
+
Layer 47: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 191 |
+
Layer 48: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 192 |
+
Layer 49: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 193 |
+
Layer 50: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 194 |
+
Layer 51: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 195 |
+
Layer 52: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 196 |
+
Layer 53: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 197 |
+
Layer 54: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 198 |
+
Layer 55: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 199 |
+
Layer 56: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 200 |
+
Layer 57: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 201 |
+
Layer 58: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 202 |
+
Layer 59: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 203 |
+
Layer 60: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 204 |
+
Layer 61: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 205 |
+
Layer 62: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 206 |
+
Layer 63: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 207 |
+
Layer 64: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 208 |
+
Layer 65: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 209 |
+
Layer 66: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 210 |
+
Layer 67: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 211 |
+
Layer 68: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 212 |
+
Layer 69: 1387.19, 8.00, 1395.19 52.00 MiB
|
| 213 |
+
Layer 70: 321.56, 0.00, 321.56 MiB (output layer)
|
| 214 |
+
--------------------------------------------------------------------------
|
| 215 |
+
Total : 93363.29, 560.00, 93923.29 MiB
|
| 216 |
+
Memory required for model tensors + cache: 94245 MiB
|
| 217 |
+
Memory available on all devices - compute: 11206 MiB
|
| 218 |
+
llm_load_tensors: ggml ctx size = 1.02 MiB
|
| 219 |
+
llm_load_tensors: offloading 70 repeating layers to GPU
|
| 220 |
+
llm_load_tensors: offloading non-repeating layers to GPU
|
| 221 |
+
llm_load_tensors: offloaded 71/71 layers to GPU
|
| 222 |
+
llm_load_tensors: CPU buffer size = 321.56 MiB
|
| 223 |
+
llm_load_tensors: CUDA0 buffer size = 93684.87 MiB
|
| 224 |
+
....................................................................................................
|
| 225 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 226 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 227 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 228 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 229 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 230 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 231 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 232 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 233 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 234 |
+
warning: madvise(..., MADV_DONTNEED) failed: Cannot allocate memory
|
| 235 |
+
llm_load_tensors: dense parameters loaded in 125.48s (5.44 GiB), expert parameters deferred (86.37 GiB)
|
| 236 |
+
llama_init_from_model: n_ctx = 2048
|
| 237 |
+
llama_init_from_model: n_batch = 256
|
| 238 |
+
llama_init_from_model: n_ubatch = 128
|
| 239 |
+
llama_init_from_model: flash_attn = 1
|
| 240 |
+
llama_init_from_model: attn_max_b = 0
|
| 241 |
+
llama_init_from_model: fused_moe = 1
|
| 242 |
+
llama_init_from_model: grouped er = 0
|
| 243 |
+
llama_init_from_model: fused_up_gate = 1
|
| 244 |
+
llama_init_from_model: fused_mmad = 1
|
| 245 |
+
llama_init_from_model: rope_cache = 0
|
| 246 |
+
llama_init_from_model: graph_reuse = 1
|
| 247 |
+
llama_init_from_model: k_cache_hadam = 0
|
| 248 |
+
llama_init_from_model: v_cache_hadam = 0
|
| 249 |
+
llama_init_from_model: split_mode_graph_scheduling = 0
|
| 250 |
+
llama_init_from_model: reduce_type = f16
|
| 251 |
+
llama_init_from_model: sched_async = 0
|
| 252 |
+
llama_init_from_model: ser = -1, 0
|
| 253 |
+
llama_init_from_model: freq_base = 500000.0
|
| 254 |
+
llama_init_from_model: freq_scale = 0.015625
|
| 255 |
+
llama_kv_cache_init: CUDA0 KV buffer size = 560.00 MiB
|
| 256 |
+
llama_init_from_model: KV self size = 560.00 MiB, K (f16): 280.00 MiB, V (f16): 280.00 MiB
|
| 257 |
+
llama_init_from_model: CUDA_Host output buffer size = 0.38 MiB
|
| 258 |
+
llama_init_from_model: CUDA0 compute buffer size = 51.00 MiB
|
| 259 |
+
llama_init_from_model: CUDA_Host compute buffer size = 2.50 MiB
|
| 260 |
+
llama_init_from_model: graph nodes = 3174
|
| 261 |
+
llama_init_from_model: graph splits = 2
|
| 262 |
+
llama_init_from_model: enabling only_active_experts scheduling
|
| 263 |
+
|
| 264 |
+
system_info: n_threads = 20 / 20 | AVX = 0 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 0 | NEON = 1 | SVE = 0 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | MATMUL_INT8 = 0 |
|
| 265 |
+
sampling:
|
| 266 |
+
repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
|
| 267 |
+
top_k = 40, tfs_z = 1.000, top_p = 0.950, min_p = 0.050, typical_p = 1.000, temp = 0.200
|
| 268 |
+
mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
|
| 269 |
+
xtc_probability = 0.000, xtc_threshold = 1.000, top_n_sigma = 0.000
|
| 270 |
+
adaptive_target = -1.00, adaptive_decay = 0.90
|
| 271 |
+
sampling order:
|
| 272 |
+
CFG -> Penalties -> dry -> top_k -> tfs_z -> typical_p -> top_p -> min_p -> xtc -> top_n_sigma -> temperature -> adaptive_p
|
| 273 |
+
generate: n_ctx = 2048, n_batch = 256, n_predict = 48, n_keep = 1
|
| 274 |
+
|
| 275 |
+
|
| 276 |
+
Write a tiny Python function that computes Fibonacci numbers iteratively. Your function should be named 'fibonacci' and take a single integer argument 'n'. It should return the nth Fibonacci number. The function should handle edge cases where n is 0 or 1.
|
| 277 |
+
|
| 278 |
+
We are computing the Fibonacci sequence iteratively
|
| 279 |
+
llama_print_timings: load time = 126743.48 ms
|
| 280 |
+
llama_print_timings: sample time = 1.94 ms / 48 runs ( 0.04 ms per token, 24742.27 tokens per second)
|
| 281 |
+
llama_print_timings: prompt eval time = 952.45 ms / 12 tokens ( 79.37 ms per token, 12.60 tokens per second)
|
| 282 |
+
llama_print_timings: eval time = 3427.81 ms / 47 runs ( 72.93 ms per token, 13.71 tokens per second)
|
| 283 |
+
llama_print_timings: total time = 130246.48 ms / 59 tokens
|
| 284 |
+
~ggml_backend_cuda_context: have 2 graphs
|
| 285 |
+
Log end
|
MANIFEST.tsv
CHANGED
|
@@ -1,18 +1,21 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
IQ3_S/eval-offload.tsv 192
|
| 2 |
+
IQ3_S/eval-ppl.log 18261
|
| 3 |
+
IQ3_S/eval-ppl.ngl99.log 18261
|
| 4 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00001-of-00010.gguf 11377052512
|
| 5 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00002-of-00010.gguf 10150217440
|
| 6 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00003-of-00010.gguf 10150217440
|
| 7 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00004-of-00010.gguf 10150217440
|
| 8 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00005-of-00010.gguf 10150217440
|
| 9 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00006-of-00010.gguf 10150217440
|
| 10 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00007-of-00010.gguf 10150217440
|
| 11 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00008-of-00010.gguf 10150217440
|
| 12 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00009-of-00010.gguf 10150217440
|
| 13 |
+
IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00010-of-00010.gguf 5997855808
|
| 14 |
+
IQ3_S/quantize-command.txt 808
|
| 15 |
+
IQ3_S/quantize.log 172866
|
| 16 |
+
IQ3_S/result.tsv 341
|
| 17 |
+
IQ3_S/SHA256SUMS 1210
|
| 18 |
+
IQ3_S/smoke.log 18802
|
| 19 |
+
IQ3_S/smoke.ngl99.log 18802
|
| 20 |
+
README.md 5631
|
| 21 |
+
results.tsv 1345
|
README.md
CHANGED
|
@@ -26,7 +26,7 @@ Until Laguna support lands in the runners you use, these files should be treated
|
|
| 26 |
| `IQ3_K_R4` | `ik-spark-core` | finished | 91.81 GiB | 2.4770 +/- 0.04102 | pending | Better quality while still on the ik CUDA fast path. |
|
| 27 |
| `IQ4_KS_R4` | `ik-spark-core` | finished | 113.20 GiB | 2.4140 +/- 0.03950 | pending | Strong quality while still optimized for ik CUDA. |
|
| 28 |
| `IQ3_S` | `vanilla-core` | finished | 91.81 GiB | 2.4683 +/- 0.04051 | pending | Vanilla CUDA MMQ-supported imatrix low-bit baseline. |
|
| 29 |
-
| `Q4_K_M` | `vanilla-core` |
|
| 30 |
| `IQ4_XS` | `vanilla-core` | pending | pending | pending | pending | Vanilla CUDA-supported nonlinear 4-bit-ish candidate. |
|
| 31 |
| `Q5_K_M` | `vanilla-core` | pending | pending | pending | pending | Higher-quality broadly compatible reference. |
|
| 32 |
| `Q6_K` | `vanilla-optional` | pending | pending | pending | pending | Optional near-reference quant if pool/upload budget allows. |
|
|
|
|
| 26 |
| `IQ3_K_R4` | `ik-spark-core` | finished | 91.81 GiB | 2.4770 +/- 0.04102 | pending | Better quality while still on the ik CUDA fast path. |
|
| 27 |
| `IQ4_KS_R4` | `ik-spark-core` | finished | 113.20 GiB | 2.4140 +/- 0.03950 | pending | Strong quality while still optimized for ik CUDA. |
|
| 28 |
| `IQ3_S` | `vanilla-core` | finished | 91.81 GiB | 2.4683 +/- 0.04051 | pending | Vanilla CUDA MMQ-supported imatrix low-bit baseline. |
|
| 29 |
+
| `Q4_K_M` | `vanilla-core` | finished | 127.33 GiB | 2.4084 +/- 0.03947 | pending | Broadly compatible CUDA-supported 4-bit-ish baseline. |
|
| 30 |
| `IQ4_XS` | `vanilla-core` | pending | pending | pending | pending | Vanilla CUDA-supported nonlinear 4-bit-ish candidate. |
|
| 31 |
| `Q5_K_M` | `vanilla-core` | pending | pending | pending | pending | Higher-quality broadly compatible reference. |
|
| 32 |
| `Q6_K` | `vanilla-optional` | pending | pending | pending | pending | Optional near-reference quant if pool/upload budget allows. |
|
results.tsv
CHANGED
|
@@ -2,4 +2,5 @@ quant lane status files bytes sha256sums ppl ppl_unc kld kld_unc kl_reference co
|
|
| 2 |
IQ3_K_R4 ik-spark-core finished 10 98576647840 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_K_R4/SHA256SUMS 2.4770 0.04102 pending pending pending 2026-06-21T01:43:29-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_K_R4/quantize.log
|
| 3 |
IQ3_S vanilla-core finished 10 98576647840 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/SHA256SUMS 2.4683 0.04051 pending pending pending 2026-06-21T02:40:21-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/quantize.log
|
| 4 |
IQ4_KS_R4 ik-spark-core finished 10 121548351136 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ4_KS_R4/SHA256SUMS 2.4140 0.03950 pending pending pending 2026-06-21T01:14:37-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ4_KS_R4/quantize.log
|
|
|
|
| 5 |
IQ2_K_R4 ik-spark-core finished 10 69096687264 /mnt/pool/gguf/laguna-m1/quants/public-v1/IQ2_K_R4/SHA256SUMS 2.6686 0.04561 pending pending pending 2026-06-20T22:58:57-04:00 /mnt/pool/gguf/laguna-m1/quants/public-v1/IQ2_K_R4/quantize.log
|
|
|
|
| 2 |
IQ3_K_R4 ik-spark-core finished 10 98576647840 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_K_R4/SHA256SUMS 2.4770 0.04102 pending pending pending 2026-06-21T01:43:29-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_K_R4/quantize.log
|
| 3 |
IQ3_S vanilla-core finished 10 98576647840 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/SHA256SUMS 2.4683 0.04051 pending pending pending 2026-06-21T02:40:21-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ3_S/quantize.log
|
| 4 |
IQ4_KS_R4 ik-spark-core finished 10 121548351136 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ4_KS_R4/SHA256SUMS 2.4140 0.03950 pending pending pending 2026-06-21T01:14:37-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ4_KS_R4/quantize.log
|
| 5 |
+
Q4_K_M vanilla-core finished 10 136723580576 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/Q4_K_M/SHA256SUMS 2.4084 0.03947 pending pending pending 2026-06-21T04:20:35-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/Q4_K_M/quantize.log
|
| 6 |
IQ2_K_R4 ik-spark-core finished 10 69096687264 /mnt/pool/gguf/laguna-m1/quants/public-v1/IQ2_K_R4/SHA256SUMS 2.6686 0.04561 pending pending pending 2026-06-20T22:58:57-04:00 /mnt/pool/gguf/laguna-m1/quants/public-v1/IQ2_K_R4/quantize.log
|