Add chat template block (runtime-owned chat templating)

v79/lfm2-5-230m-2048.json

@@ -1,20 +1,68 @@
-{
-  "schema_version": 1,
-  "_comment": "lfm2-5-230m-2048 - Qualcomm Hexagon v79 (QHexRT) deployment manifest.",
-  "model": { "name": "lfm2-5-230m-2048", "family": "llm", "dsp_arch": "v79" },
-  "params": {
-    "hidden": 1024, "vocab": 65536, "n_layers": 14, "max_ctx": 2048,
-    "kv_dim": 512, "head_dim": 64, "rope_theta": 1000000.0, "eos_token_id": 7
-  },
-  "artifacts": {
-    "contexts": { "decode": { "bin": "lfm230_dec_2048_w8.bin" }, "prefill": { "bin": "lfm230_pf_2048_w8.bin" }, "lmhead": { "bin": "lfm230_lmh_w8.bin" } },
-    "embed": "lfm_embed_f16.bin",
-    "tokenizer": "tokenizer.json"
-  },
-  "plan": {
-    "steps": [
-      { "host": "lfm_generate", "params": { "prefill": "lfm230_pf_2048_w8", "decode": "lfm230_dec_2048_w8", "lmhead": "lfm230_lmh_w8", "attn_idx": "2,4,6,8,10,12", "conv_idx": "0,1,3,5,7,9,11,13" } },
-      { "emit": "all" }
-    ]
-  }
-}
+{
+  "schema_version": 1,
+  "_comment": "lfm2-5-230m-2048 - Qualcomm Hexagon v79 (QHexRT) deployment manifest.",
+  "model": {
+    "name": "lfm2-5-230m-2048",
+    "family": "llm",
+    "dsp_arch": "v79"
+  },
+  "chat": {
+    "bos": [
+      1
+    ],
+    "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",
+      "suffix": "<|im_end|>\n"
+    }
+  },
+  "params": {
+    "hidden": 1024,
+    "vocab": 65536,
+    "n_layers": 14,
+    "max_ctx": 2048,
+    "kv_dim": 512,
+    "head_dim": 64,
+    "rope_theta": 1000000.0,
+    "eos_token_id": 7
+  },
+  "artifacts": {
+    "contexts": {
+      "decode": {
+        "bin": "lfm230_dec_2048_w8.bin"
+      },
+      "prefill": {
+        "bin": "lfm230_pf_2048_w8.bin"
+      },
+      "lmhead": {
+        "bin": "lfm230_lmh_w8.bin"
+      }
+    },
+    "embed": "lfm_embed_f16.bin",
+    "tokenizer": "tokenizer.json"
+  },
+  "plan": {
+    "steps": [
+      {
+        "host": "lfm_generate",
+        "params": {
+          "prefill": "lfm230_pf_2048_w8",
+          "decode": "lfm230_dec_2048_w8",
+          "lmhead": "lfm230_lmh_w8",
+          "attn_idx": "2,4,6,8,10,12",
+          "conv_idx": "0,1,3,5,7,9,11,13"
+        }
+      },
+      {
+        "emit": "all"
+      }
+    ]
+  }
+}