v75 (S24/SM8650, soc 57): Qwen3-0.6B W8-decode + fp16-prefill bundle — Plane A 6/6

.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
+v75/tokenizer.json filter=lfs diff=lfs merge=lfs -text

v75/qwen06_decn1k_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf5b001ede6b08880eaae081618eb120f76a0cad0ead6cbec15ff0b30d41cb3e
+size 445235200

v75/qwen06_embed_f16.bin

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

v75/qwen06_lmh_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a4ed98b229a6995675c8ff02358a211c4e2f6654e6b0680be7bd1c88f62183e
+size 157859840

v75/qwen06_pf_f16.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:334332b2ff4263514b2cad6bdaefe64c4e12fe19559c2e8ef1ac3d63a07c7986
+size 895619072

v75/qwen3-0.6b-1024final.json

@@ -0,0 +1,49 @@
+{
+  "schema_version": 1,
+  "_comment": "Standalone Qwen3-0.6B as a first-class native QHexRT LLM (NO Genie). Same architecture as InternVL3.5-1B's LLM backbone (16q/8kv, head_dim 128, QK-norm, standard 1D RoPE theta 1e6, tied lm-head), which is the proven 16-head materialized attention that runs COHERENTLY at MAXCTX=2048 on v79 (unlike Llama's 32-head). decode graph qwen06_dec (fp16, 2048) + NPU tied lm-head (qwen06_lmh = x@E.T) + host embed lookup, full-dim 1D RoPE, cmask, sliding-window KV ring, driven by the qwen3_generate host-op. Decode-over-prompt seed (no batched prefill yet).",
+  "model": {
+    "name": "qwen3-0.6b-1024final",
+    "family": "llm",
+    "dsp_arch": "v75",
+    "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"
+  },
+  "plan": {
+    "steps": [
+      {
+        "host": "qwen3_generate",
+        "params": {
+          "prefill": "qwen06_pf_f16",
+          "decode": "qwen06_decn1k_w8",
+          "lmhead": "qwen06_lmh_w8",
+          "max_new": 64,
+          "neg": -50000
+        }
+      }
+    ]
+  }
+}

v75/tokenizer.json

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