Upload v79/internvl3_5-1b.json with huggingface_hub

v79/internvl3_5-1b.json

@@ -0,0 +1,28 @@
+{
+  "schema_version": 1,
+  "_comment": "InternVL3.5-1B VLM (InternViT-300M + Qwen3-0.6B) fully on Hexagon v79 via the QHexRT plan-interpreter. The internvl_generate host-op runs the whole pipeline through QHexRT: host preproc/patch-embed -> NPU vision encoder -> host pixel-shuffle/projector -> host tokenize+splice@IMG_CONTEXT -> NPU prefill -> resident-ion-KV decode loop + in-graph int8 lm-head. W4A16 decode. Fixed prompt 'Describe this image in detail' (S=310).",
+  "model": { "name": "internvl3_5-1b", "family": "vlm", "dsp_arch": "v79" },
+  "params": {
+    "hidden": 1024, "vocab": 151936, "n_layers": 28, "max_ctx": 512,
+    "kv_dim": 1024, "head_dim": 128, "rope_theta": 1000000.0,
+    "vision_px": 448, "eos_token_id": 151645
+  },
+  "artifacts": {
+    "contexts": {
+      "vision":  { "bin": "ivl_enc2.bin" },
+      "prefill": { "bin": "ivl_pf.bin" },
+      "decode":  { "bin": "ivl_dec_w4.bin" },
+      "lmhead":  { "bin": "ivl_lmh_q.bin" }
+    },
+    "embed": "ivl_embed_f16.bin",
+    "tokenizer": "tokenizer.json",
+    "fixture_dir": "vlm"
+  },
+  "plan": {
+    "steps": [
+      { "host": "internvl_generate",
+        "params": { "vision": "ivl_enc2", "prefill": "ivl_pf", "decode": "ivl_dec_w4", "lmhead": "ivl_lmh_q" } },
+      { "emit": "all" }
+    ]
+  }
+}