Hugging Face's logo

sigargv
/
Laguna-M.1-GGUF

Text Generation
GGUF
llama.cpp
ik_llama.cpp
laguna
laguna-m.1
Mixture of Experts
imatrix
poolside
conversational
Model card Files
xet
 Community
sigargv commited on
Commit
6aa57bc
Β·
verified Β·
1 Parent(s): 3943f51

Add files using upload-large-folder tool

Browse files
Files changed (20) hide show
  1. .gitattributes +10 -0
  2. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00001-of-00010.gguf +3 -0
  3. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00002-of-00010.gguf +3 -0
  4. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00003-of-00010.gguf +3 -0
  5. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00004-of-00010.gguf +3 -0
  6. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00005-of-00010.gguf +3 -0
  7. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00006-of-00010.gguf +3 -0
  8. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00007-of-00010.gguf +3 -0
  9. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00008-of-00010.gguf +3 -0
  10. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00009-of-00010.gguf +3 -0
  11. IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00010-of-00010.gguf +3 -0
  12. IQ4_KS_R4/SHA256SUMS +10 -0
  13. IQ4_KS_R4/eval-ppl.log +276 -0
  14. IQ4_KS_R4/quantize-command.txt +8 -0
  15. IQ4_KS_R4/quantize.log +0 -0
  16. IQ4_KS_R4/result.tsv +2 -0
  17. IQ4_KS_R4/smoke.log +290 -0
  18. MANIFEST.tsv +18 -21
  19. README.md +1 -1
  20. results.tsv +1 -0
.gitattributes CHANGED
@@ -63,3 +63,13 @@ IQ3_S/Laguna-M.1-IQ3_S-imatrix-public-v1-00008-of-00010.gguf filter=lfs diff=lfs
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
 
 
 
 
 
 
 
 
 
 
 
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
66
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00009-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
67
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00002-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
68
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00001-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
69
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00004-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
70
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00007-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
71
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00005-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
72
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00008-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
73
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00003-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
74
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00006-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
75
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00010-of-00010.gguf filter=lfs diff=lfs merge=lfs -text
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00001-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c7d82204ba7372320dfe5e5be87ddf6341727248733f544143b109e8f62a4d2e
3
+ size 13341584224
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00002-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f51d35c7df512e1437aa5c217a1e6494a29147375c1f1e3de2842c7d727e873
3
+ size 12597593824
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00003-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52ed22ce8246e6e5f52022dd7121f4b3e45b4b30991be5849cdb8ca449cb2e0b
3
+ size 12594448096
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00004-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b1516d29b415e478d52d2bfccc116105d9d5b54755d78ccf78f7c8871b28d960
3
+ size 12594448096
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00005-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e84068e42403fe9f9cb4c3377880cbc92f1b57d97553a87bc84f998ca68850f6
3
+ size 12597593824
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00006-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4804cf81ab97595e30760af0125795b91194a10644c9998158aa58cfda2ee9b8
3
+ size 12594448096
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00007-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9956dfb8e5afb9b1f74f65ca107d0250d5a584b4c015ba7a989e5713191cbe84
3
+ size 12594448096
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00008-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:887732b61a501109fe709f146edc73c7c78a343b8a74a7a8169d9cf403938d05
3
+ size 12597593824
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00009-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3bdefcb2bc571b3cdf5c83ecaf36bdcd871754e586c934dbcd33c3afe9478c3d
3
+ size 12594448096
IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00010-of-00010.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca356e9a13e87476790a512d5d00e48a5bd4ab408a3cbe3265e4424bae0ba2d8
3
+ size 7441744960
IQ4_KS_R4/SHA256SUMS ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ c7d82204ba7372320dfe5e5be87ddf6341727248733f544143b109e8f62a4d2e Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00001-of-00010.gguf
2
+ 1f51d35c7df512e1437aa5c217a1e6494a29147375c1f1e3de2842c7d727e873 Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00002-of-00010.gguf
3
+ 52ed22ce8246e6e5f52022dd7121f4b3e45b4b30991be5849cdb8ca449cb2e0b Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00003-of-00010.gguf
4
+ b1516d29b415e478d52d2bfccc116105d9d5b54755d78ccf78f7c8871b28d960 Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00004-of-00010.gguf
5
+ e84068e42403fe9f9cb4c3377880cbc92f1b57d97553a87bc84f998ca68850f6 Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00005-of-00010.gguf
6
+ 4804cf81ab97595e30760af0125795b91194a10644c9998158aa58cfda2ee9b8 Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00006-of-00010.gguf
7
+ 9956dfb8e5afb9b1f74f65ca107d0250d5a584b4c015ba7a989e5713191cbe84 Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00007-of-00010.gguf
8
+ 887732b61a501109fe709f146edc73c7c78a343b8a74a7a8169d9cf403938d05 Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00008-of-00010.gguf
9
+ 3bdefcb2bc571b3cdf5c83ecaf36bdcd871754e586c934dbcd33c3afe9478c3d Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00009-of-00010.gguf
10
+ ca356e9a13e87476790a512d5d00e48a5bd4ab408a3cbe3265e4424bae0ba2d8 Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00010-of-00010.gguf
IQ4_KS_R4/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 - 118148 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/IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-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 = 345
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€ˆ|UNK|〉", "γ€ˆ|CODE_START|〉",...
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 iq4_ks_r4: 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€ˆ|EOS|〉')
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 = IQ4_KS_R4 - 4.25 bpw
122
+ llm_load_print_meta: model params = 225.796 B
123
+ llm_load_print_meta: model size = 113.197 GiB (4.306 BPW)
124
+ llm_load_print_meta: repeating layers = 112.569 GiB (4.298 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€ˆ|EOS|〉'
130
+ print_info: EOS token = 2 'γ€ˆ|EOS|〉'
131
+ print_info: EOT token = 24 '</assistant>'
132
+ print_info: UNK token = 0 'γ€ˆ|UNK|〉'
133
+ print_info: SEP token = 8 'γ€ˆ|SEP|〉'
134
+ print_info: PAD token = 9 'γ€ˆ|PAD|〉'
135
+ print_info: MASK token = 12 'γ€ˆ|MASK|〉'
136
+ print_info: LF token = 268 'Ċ'
137
+ print_info: EOG token = 2 'γ€ˆ|EOS|〉'
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: 166.74, 16.00, 182.74 22.00 MiB
143
+ Layer 1: 166.74, 16.00, 182.74 22.00 MiB
144
+ Layer 2: 166.74, 16.00, 182.74 22.00 MiB
145
+ Layer 3: 1712.99, 16.00, 1728.99 52.00 MiB
146
+ Layer 4: 1712.99, 16.00, 1728.99 52.00 MiB
147
+ Layer 5: 1712.99, 16.00, 1728.99 52.00 MiB
148
+ Layer 6: 1712.99, 16.00, 1728.99 52.00 MiB
149
+ Layer 7: 1712.99, 16.00, 1728.99 52.00 MiB
150
+ Layer 8: 1712.99, 16.00, 1728.99 52.00 MiB
151
+ Layer 9: 1712.99, 16.00, 1728.99 52.00 MiB
152
+ Layer 10: 1712.99, 16.00, 1728.99 52.00 MiB
153
+ Layer 11: 1712.99, 16.00, 1728.99 52.00 MiB
154
+ Layer 12: 1712.99, 16.00, 1728.99 52.00 MiB
155
+ Layer 13: 1712.99, 16.00, 1728.99 52.00 MiB
156
+ Layer 14: 1712.99, 16.00, 1728.99 52.00 MiB
157
+ Layer 15: 1712.99, 16.00, 1728.99 52.00 MiB
158
+ Layer 16: 1712.99, 16.00, 1728.99 52.00 MiB
159
+ Layer 17: 1712.99, 16.00, 1728.99 52.00 MiB
160
+ Layer 18: 1712.99, 16.00, 1728.99 52.00 MiB
161
+ Layer 19: 1712.99, 16.00, 1728.99 52.00 MiB
162
+ Layer 20: 1712.99, 16.00, 1728.99 52.00 MiB
163
+ Layer 21: 1712.99, 16.00, 1728.99 52.00 MiB
164
+ Layer 22: 1712.99, 16.00, 1728.99 52.00 MiB
165
+ Layer 23: 1712.99, 16.00, 1728.99 52.00 MiB
166
+ Layer 24: 1712.99, 16.00, 1728.99 52.00 MiB
167
+ Layer 25: 1712.99, 16.00, 1728.99 52.00 MiB
168
+ Layer 26: 1712.99, 16.00, 1728.99 52.00 MiB
169
+ Layer 27: 1712.99, 16.00, 1728.99 52.00 MiB
170
+ Layer 28: 1712.99, 16.00, 1728.99 52.00 MiB
171
+ Layer 29: 1712.99, 16.00, 1728.99 52.00 MiB
172
+ Layer 30: 1712.99, 16.00, 1728.99 52.00 MiB
173
+ Layer 31: 1712.99, 16.00, 1728.99 52.00 MiB
174
+ Layer 32: 1712.99, 16.00, 1728.99 52.00 MiB
175
+ Layer 33: 1712.99, 16.00, 1728.99 52.00 MiB
176
+ Layer 34: 1712.99, 16.00, 1728.99 52.00 MiB
177
+ Layer 35: 1712.99, 16.00, 1728.99 52.00 MiB
178
+ Layer 36: 1712.99, 16.00, 1728.99 52.00 MiB
179
+ Layer 37: 1712.99, 16.00, 1728.99 52.00 MiB
180
+ Layer 38: 1712.99, 16.00, 1728.99 52.00 MiB
181
+ Layer 39: 1712.99, 16.00, 1728.99 52.00 MiB
182
+ Layer 40: 1712.99, 16.00, 1728.99 52.00 MiB
183
+ Layer 41: 1712.99, 16.00, 1728.99 52.00 MiB
184
+ Layer 42: 1712.99, 16.00, 1728.99 52.00 MiB
185
+ Layer 43: 1712.99, 16.00, 1728.99 52.00 MiB
186
+ Layer 44: 1712.99, 16.00, 1728.99 52.00 MiB
187
+ Layer 45: 1712.99, 16.00, 1728.99 52.00 MiB
188
+ Layer 46: 1712.99, 16.00, 1728.99 52.00 MiB
189
+ Layer 47: 1712.99, 16.00, 1728.99 52.00 MiB
190
+ Layer 48: 1712.99, 16.00, 1728.99 52.00 MiB
191
+ Layer 49: 1712.99, 16.00, 1728.99 52.00 MiB
192
+ Layer 50: 1712.99, 16.00, 1728.99 52.00 MiB
193
+ Layer 51: 1712.99, 16.00, 1728.99 52.00 MiB
194
+ Layer 52: 1712.99, 16.00, 1728.99 52.00 MiB
195
+ Layer 53: 1712.99, 16.00, 1728.99 52.00 MiB
196
+ Layer 54: 1712.99, 16.00, 1728.99 52.00 MiB
197
+ Layer 55: 1712.99, 16.00, 1728.99 52.00 MiB
198
+ Layer 56: 1712.99, 16.00, 1728.99 52.00 MiB
199
+ Layer 57: 1712.99, 16.00, 1728.99 52.00 MiB
200
+ Layer 58: 1712.99, 16.00, 1728.99 52.00 MiB
201
+ Layer 59: 1712.99, 16.00, 1728.99 52.00 MiB
202
+ Layer 60: 1712.99, 16.00, 1728.99 52.00 MiB
203
+ Layer 61: 1712.99, 16.00, 1728.99 52.00 MiB
204
+ Layer 62: 1712.99, 16.00, 1728.99 52.00 MiB
205
+ Layer 63: 1712.99, 16.00, 1728.99 52.00 MiB
206
+ Layer 64: 1712.99, 16.00, 1728.99 52.00 MiB
207
+ Layer 65: 1712.99, 16.00, 1728.99 52.00 MiB
208
+ Layer 66: 1712.99, 16.00, 1728.99 52.00 MiB
209
+ Layer 67: 1712.99, 16.00, 1728.99 52.00 MiB
210
+ Layer 68: 1712.99, 16.00, 1728.99 52.00 MiB
211
+ Layer 69: 1712.99, 16.00, 1728.99 52.00 MiB
212
+ Layer 70: 321.56, 0.00, 321.56 MiB (output layer)
213
+ --------------------------------------------------------------------------
214
+ Total : 115270.81, 1120.00, 116390.81 MiB
215
+ Memory required for model tensors + cache: 116712 MiB
216
+ Memory available on all devices - compute: 116993 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 = 115592.39 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 306.54s (6.02 GiB), expert parameters deferred (107.17 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 1968.96 ms
266
+ perplexity: calculating perplexity over 8 chunks, n_ctx=4096, batch_size=512, n_seq=1
267
+ perplexity: 25.74 seconds per pass - ETA 3.42 minutes
268
+ [1]1.3464,[2]2.1489,[3]1.7387,[4]2.3562,[5]2.5638,[6]2.6082,[7]2.5614,[8]2.4140,
269
+ llama_print_timings: load time = 325675.85 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 = 207940.29 ms / 32768 tokens ( 6.35 ms per token, 157.58 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 = 527491.35 ms / 32769 tokens
274
+ ~ggml_backend_cuda_context: have 1 graphs
275
+
276
+ Final estimate: PPL over 8 chunks for n_ctx=4096 = 2.4140 +/- 0.03950
IQ4_KS_R4/quantize-command.txt ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ quant=IQ4_KS_R4
2
+ lane=ik-spark-core
3
+ started_at=2026-06-20T23:58:03-04:00
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/IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1.gguf IQ4_KS_R4 20
IQ4_KS_R4/quantize.log ADDED
The diff for this file is too large to render. See raw diff
 
IQ4_KS_R4/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
+ 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
IQ4_KS_R4/smoke.log ADDED
@@ -0,0 +1,290 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 - 32682 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/IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-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 = 345
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€ˆ|UNK|〉", "γ€ˆ|CODE_START|〉",...
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 iq4_ks_r4: 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€ˆ|EOS|〉')
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 = IQ4_KS_R4 - 4.25 bpw
123
+ llm_load_print_meta: model params = 225.796 B
124
+ llm_load_print_meta: model size = 113.197 GiB (4.306 BPW)
125
+ llm_load_print_meta: repeating layers = 112.569 GiB (4.298 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€ˆ|EOS|〉'
131
+ print_info: EOS token = 2 'γ€ˆ|EOS|〉'
132
+ print_info: EOT token = 24 '</assistant>'
133
+ print_info: UNK token = 0 'γ€ˆ|UNK|〉'
134
+ print_info: SEP token = 8 'γ€ˆ|SEP|〉'
135
+ print_info: PAD token = 9 'γ€ˆ|PAD|〉'
136
+ print_info: MASK token = 12 'γ€ˆ|MASK|〉'
137
+ print_info: LF token = 268 'Ċ'
138
+ print_info: EOG token = 2 'γ€ˆ|EOS|〉'
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: 166.74, 8.00, 174.74 22.00 MiB
144
+ Layer 1: 166.74, 8.00, 174.74 22.00 MiB
145
+ Layer 2: 166.74, 8.00, 174.74 22.00 MiB
146
+ Layer 3: 1712.99, 8.00, 1720.99 52.00 MiB
147
+ Layer 4: 1712.99, 8.00, 1720.99 52.00 MiB
148
+ Layer 5: 1712.99, 8.00, 1720.99 52.00 MiB
149
+ Layer 6: 1712.99, 8.00, 1720.99 52.00 MiB
150
+ Layer 7: 1712.99, 8.00, 1720.99 52.00 MiB
151
+ Layer 8: 1712.99, 8.00, 1720.99 52.00 MiB
152
+ Layer 9: 1712.99, 8.00, 1720.99 52.00 MiB
153
+ Layer 10: 1712.99, 8.00, 1720.99 52.00 MiB
154
+ Layer 11: 1712.99, 8.00, 1720.99 52.00 MiB
155
+ Layer 12: 1712.99, 8.00, 1720.99 52.00 MiB
156
+ Layer 13: 1712.99, 8.00, 1720.99 52.00 MiB
157
+ Layer 14: 1712.99, 8.00, 1720.99 52.00 MiB
158
+ Layer 15: 1712.99, 8.00, 1720.99 52.00 MiB
159
+ Layer 16: 1712.99, 8.00, 1720.99 52.00 MiB
160
+ Layer 17: 1712.99, 8.00, 1720.99 52.00 MiB
161
+ Layer 18: 1712.99, 8.00, 1720.99 52.00 MiB
162
+ Layer 19: 1712.99, 8.00, 1720.99 52.00 MiB
163
+ Layer 20: 1712.99, 8.00, 1720.99 52.00 MiB
164
+ Layer 21: 1712.99, 8.00, 1720.99 52.00 MiB
165
+ Layer 22: 1712.99, 8.00, 1720.99 52.00 MiB
166
+ Layer 23: 1712.99, 8.00, 1720.99 52.00 MiB
167
+ Layer 24: 1712.99, 8.00, 1720.99 52.00 MiB
168
+ Layer 25: 1712.99, 8.00, 1720.99 52.00 MiB
169
+ Layer 26: 1712.99, 8.00, 1720.99 52.00 MiB
170
+ Layer 27: 1712.99, 8.00, 1720.99 52.00 MiB
171
+ Layer 28: 1712.99, 8.00, 1720.99 52.00 MiB
172
+ Layer 29: 1712.99, 8.00, 1720.99 52.00 MiB
173
+ Layer 30: 1712.99, 8.00, 1720.99 52.00 MiB
174
+ Layer 31: 1712.99, 8.00, 1720.99 52.00 MiB
175
+ Layer 32: 1712.99, 8.00, 1720.99 52.00 MiB
176
+ Layer 33: 1712.99, 8.00, 1720.99 52.00 MiB
177
+ Layer 34: 1712.99, 8.00, 1720.99 52.00 MiB
178
+ Layer 35: 1712.99, 8.00, 1720.99 52.00 MiB
179
+ Layer 36: 1712.99, 8.00, 1720.99 52.00 MiB
180
+ Layer 37: 1712.99, 8.00, 1720.99 52.00 MiB
181
+ Layer 38: 1712.99, 8.00, 1720.99 52.00 MiB
182
+ Layer 39: 1712.99, 8.00, 1720.99 52.00 MiB
183
+ Layer 40: 1712.99, 8.00, 1720.99 52.00 MiB
184
+ Layer 41: 1712.99, 8.00, 1720.99 52.00 MiB
185
+ Layer 42: 1712.99, 8.00, 1720.99 52.00 MiB
186
+ Layer 43: 1712.99, 8.00, 1720.99 52.00 MiB
187
+ Layer 44: 1712.99, 8.00, 1720.99 52.00 MiB
188
+ Layer 45: 1712.99, 8.00, 1720.99 52.00 MiB
189
+ Layer 46: 1712.99, 8.00, 1720.99 52.00 MiB
190
+ Layer 47: 1712.99, 8.00, 1720.99 52.00 MiB
191
+ Layer 48: 1712.99, 8.00, 1720.99 52.00 MiB
192
+ Layer 49: 1712.99, 8.00, 1720.99 52.00 MiB
193
+ Layer 50: 1712.99, 8.00, 1720.99 52.00 MiB
194
+ Layer 51: 1712.99, 8.00, 1720.99 52.00 MiB
195
+ Layer 52: 1712.99, 8.00, 1720.99 52.00 MiB
196
+ Layer 53: 1712.99, 8.00, 1720.99 52.00 MiB
197
+ Layer 54: 1712.99, 8.00, 1720.99 52.00 MiB
198
+ Layer 55: 1712.99, 8.00, 1720.99 52.00 MiB
199
+ Layer 56: 1712.99, 8.00, 1720.99 52.00 MiB
200
+ Layer 57: 1712.99, 8.00, 1720.99 52.00 MiB
201
+ Layer 58: 1712.99, 8.00, 1720.99 52.00 MiB
202
+ Layer 59: 1712.99, 8.00, 1720.99 52.00 MiB
203
+ Layer 60: 1712.99, 8.00, 1720.99 52.00 MiB
204
+ Layer 61: 1712.99, 8.00, 1720.99 52.00 MiB
205
+ Layer 62: 1712.99, 8.00, 1720.99 52.00 MiB
206
+ Layer 63: 1712.99, 8.00, 1720.99 52.00 MiB
207
+ Layer 64: 1712.99, 8.00, 1720.99 52.00 MiB
208
+ Layer 65: 1712.99, 8.00, 1720.99 52.00 MiB
209
+ Layer 66: 1712.99, 8.00, 1720.99 52.00 MiB
210
+ Layer 67: 1712.99, 8.00, 1720.99 52.00 MiB
211
+ Layer 68: 1712.99, 8.00, 1720.99 52.00 MiB
212
+ Layer 69: 1712.99, 8.00, 1720.99 52.00 MiB
213
+ Layer 70: 321.56, 0.00, 321.56 MiB (output layer)
214
+ --------------------------------------------------------------------------
215
+ Total : 115270.81, 560.00, 115830.81 MiB
216
+ Memory required for model tensors + cache: 116152 MiB
217
+ Memory available on all devices - compute: 32107 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 = 115592.39 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 403.44s (6.02 GiB), expert parameters deferred (107.17 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.
277
+
278
+ I need a tiny Python function that computes Fibonacci numbers iteratively. Can you help me with that?
279
+
280
+ The Fibonacci sequence is defined as:
281
+ - F(0) = 0
282
+ - F(1) = 1
283
+ -
284
+ llama_print_timings: load time = 407199.61 ms
285
+ llama_print_timings: sample time = 1.91 ms / 48 runs ( 0.04 ms per token, 25117.74 tokens per second)
286
+ llama_print_timings: prompt eval time = 1085.88 ms / 12 tokens ( 90.49 ms per token, 11.05 tokens per second)
287
+ llama_print_timings: eval time = 3545.93 ms / 47 runs ( 75.45 ms per token, 13.25 tokens per second)
288
+ llama_print_timings: total time = 410756.82 ms / 59 tokens
289
+ ~ggml_backend_cuda_context: have 2 graphs
290
+ Log end
MANIFEST.tsv CHANGED
@@ -1,21 +1,18 @@
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
 
1
+ IQ4_KS_R4/eval-ppl.log 18283
2
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00001-of-00010.gguf 13341584224
3
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00002-of-00010.gguf 12597593824
4
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00003-of-00010.gguf 12594448096
5
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00004-of-00010.gguf 12594448096
6
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00005-of-00010.gguf 12597593824
7
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00006-of-00010.gguf 12594448096
8
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00007-of-00010.gguf 12594448096
9
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00008-of-00010.gguf 12597593824
10
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00009-of-00010.gguf 12594448096
11
+ IQ4_KS_R4/Laguna-M.1-IQ4_KS_R4-imatrix-public-v1-00010-of-00010.gguf 7441744960
12
+ IQ4_KS_R4/quantize-command.txt 825
13
+ IQ4_KS_R4/quantize.log 175083
14
+ IQ4_KS_R4/result.tsv 355
15
+ IQ4_KS_R4/SHA256SUMS 1250
16
+ IQ4_KS_R4/smoke.log 18753
17
+ README.md 5646
18
+ results.tsv 1595
 
 
 
README.md CHANGED
@@ -27,7 +27,7 @@ Until Laguna support lands in the runners you use, these files should be treated
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. |
33
  | `IQ4_K_R4` | `ik-optional` | pending | pending | pending | pending | Larger/higher-quality ik-specific CUDA fast-path candidate. |
 
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` | finished | 112.80 GiB | 2.4227 +/- 0.03976 | 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. |
33
  | `IQ4_K_R4` | `ik-optional` | pending | pending | pending | pending | Larger/higher-quality ik-specific CUDA fast-path candidate. |
results.tsv CHANGED
@@ -2,5 +2,6 @@ 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
  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
 
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
+ IQ4_XS vanilla-core finished 10 121120147104 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ4_XS/SHA256SUMS 2.4227 0.03976 pending pending pending 2026-06-21T05:08:18-04:00 /home/janet/laguna-gguf/laguna-m1/quants/public-v1/IQ4_XS/quantize.log
6
  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
7
  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