Add v79 (Hexagon v79 / SM8750, S25) bundle

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 v81/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+v79/tokenizer.json filter=lfs diff=lfs merge=lfs -text

v79/qwen06_decn1k_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff658f9eb05d12848f0437805de345bf4b48979108f8085742051c95e1c5c294
+size 445333504

v79/qwen06_embed_f16.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:616f9374ddbaae3a033a0174557d502719154882e21e5d975ab5b58e0244459d
+size 311164928

v79/qwen06_lmh_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da3a1bb48125de01d22e5d03297dda2b0fa2dc88c0296890a501a6d6c204812f
+size 157868032

v79/qwen06_pf_f16.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:baff8ecde24c94c2fbc25548c6829246d91feae738d4f4eced78c7df646e2840
+size 895508480

v79/qwen3-0.6b-1024final.json

@@ -0,0 +1,28 @@
+{
+  "schema_version": 1,
+  "_comment": "Standalone Qwen3-0.6B native QHexRT LLM (Halo brain). 16q/8kv, head_dim 128, QK-norm, 1D RoPE theta 1e6, tied lm-head; 16-head materialized attention runs coherently at MAXCTX. decode qwen06_decn1k_w8 (w8, 1024) + NPU tied lm-head (qwen06_lmh_w8) + host embed + full-dim 1D RoPE + cmask + sliding-window KV ring, driven by qwen3_generate. The `chat` block below makes templating runtime-owned (ChatML) and DISABLES Qwen3 thinking via an empty <think></think> assistant prefix (matches Halo's device-validated tool/chat prompt).",
+  "model": { "name": "qwen3-0.6b-1024final", "family": "llm", "dsp_arch": "v79", "tokenizer_pre": "qwen2" },
+  "params": {
+    "hidden": 1024, "vocab": 151936, "n_layers": 28, "max_ctx": 1024,
+    "kv_dim": 1024, "head_dim": 128, "rope_theta": 1000000.0, "eos_token_id": 151645
+  },
+  "artifacts": {
+    "contexts": {
+      "prefill": { "bin": "qwen06_pf_f16.bin" }, "decode": { "bin": "qwen06_decn1k_w8.bin" },
+      "lmhead": { "bin": "qwen06_lmh_w8.bin" }
+    },
+    "embed": "qwen06_embed_f16.bin",
+    "tokenizer": "tokenizer.json"
+  },
+  "chat": {
+    "system":    { "prefix": "<|im_start|>system\n", "suffix": "<|im_end|>\n" },
+    "user":      { "prefix": "<|im_start|>user\n", "suffix": "<|im_end|>\n" },
+    "assistant": { "prefix": "<|im_start|>assistant\n<think>\n\n</think>\n\n", "suffix": "<|im_end|>\n" }
+  },
+  "plan": {
+    "steps": [
+      { "host": "qwen3_generate",
+        "params": { "prefill": "qwen06_pf_f16", "decode": "qwen06_decn1k_w8", "lmhead": "qwen06_lmh_w8", "max_new": 64, "neg": -50000 } }
+    ]
+  }
+}

v79/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650