{
  "artifacts": [
    {
      "file": "Ovis2.5-2B.litertlm",
      "sha256": "528485444515cef66d999dfa551be5740854612cf9bfc1003ff5b9c7281afc7b",
      "size_mb": 2049.657
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "bash scripts/ship_ovis_2b.sh",
    "quantization": "vision tower int8; Qwen3-1.7B decoder int4 blockwise-32 symmetric + OCTAV; input embedding int8 (externalized section)",
    "tool": "litert-torch (litertlm-convert fast_vlm pipeline, scripts/ship_ovis_2b.sh)",
    "tool_version": "0.10.0 (editable dev checkout 115a136 + local patches)"
  },
  "cross_runtime": [],
  "delegation": null,
  "device": {
    "records": [
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu",
          "context_length": null,
          "date": "2026-08-24",
          "decode_tokens_per_s": 26.83,
          "delegated_ops": 1298,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-cl-pinned",
            "os_build": "Android 16",
            "runtime": "litert-lm",
            "runtime_version": "0.16.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "VERBOSE: Replacing 1298 out of 1298 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (prefill_128).",
            "VERBOSE: Replacing 1298 out of 1298 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 1 (prefill_512).",
            "VERBOSE: Replacing 1244 out of 1244 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 2 (decode).",
            "results block: prefill=461.75 decode=26.83 tokens/s"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "decode_tokens": 853.0,
            "init_s": 7.99207,
            "prefill_tokens": 203.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": 461.75,
          "provenance": "measured",
          "runs": true,
          "total_ops": 1298,
          "ttft_ms": 480.0
        },
        "source": "data/device_runs/0.16.0/2026-08-24/ovis2_5-2b__galaxy-s26.json"
      }
    ]
  },
  "model": {
    "family": "ovis",
    "id": "ovis2_5-2b",
    "license": "apache-2.0",
    "source_url": "https://huggingface.co/AIDC-AI/Ovis2.5-2B",
    "task": "image-text-to-text"
  },
  "pitfalls": [
    "Reasoning VLM: the model may emit a <think>...</think> block before its final answer (matches the base model) — allow enough max-tokens (>=1024) for the answer to follow.",
    "Vision-only bundle (no audio tower): bring up the engine with the vision modality only — requesting the audio tower (.all) on a bundle with no audio section fails at session creation.",
    "The dynamic-resolution Siglip2-NaViT tower does not torch.export; shipped as a static 512x512 rewrite (valid because fullatt_block_indexes=None makes the window-reorder a no-op) — static-vs-original feature corr 0.99999964.",
    "Siglip normalization ((x-0.5)/0.5) is baked into the encoder; the runtime feeds a [0,1] NHWC image resized to 512x512.",
    "Ovis's two learned image-boundary indicator embeddings are omitted (the fast_vlm path splices only the atom embeddings) — verified to stay coherent in eager."
  ],
  "schema_version": "1.2"
}