GPU-Mart / index.html

Commit History

  1. Use greedy decoding: do_sample sampling path fails on fp16 logits in transformers.js 3.8.1; remove temp instrumentation

    Verified signature
    d6c22f2
  2. Instrument adapter (temporary): expose generator + wrapper call counter

    Verified signature
    faae2b5
  3. Adapter v3: learn expected KV dtypes via try/flip/retry (session metadata unavailable in CDN ORT build)

    Verified signature
    7434dde
  4. Add KV dtype adapter: align KV-cache feed dtypes with session input metadata (handles both fp16- and fp32-resolving ORT builds)

    Verified signature
    32cb8a0
  5. Force WASM inference: WebGPU shader-f16 corrupts this fp16 export (verified on-device); WASM output verified correct. Trim max_new_tokens to 160.

    Verified signature
    35f4ea6
  6. Fix generation: remove fp16 I/O shim (config kv_cache_dtype handles it), pin transformers.js 3.8.1, refresh stale cached configs, restore WASM fallback with shader-f16 gate

    Verified signature
    379ab55
  7. Shim: only cast when session input metadata expects float16

    Verified signature
    c2f6821
  8. Add fp16 I/O shim: cast fp32 KV feeds to fp16 and fp16 logits to fp32

    Verified signature
    a1ae561

  1. Use current Transformers.js 3.x for fp16 KV cache

    Verified signature
    4e9fd49
  2. Require WebGPU for fp16 GPU Mart inference

    Verified signature
    92cdbd8

  1. Use fp16 ONNX with graph optimizations disabled

    Verified signature
    1021060
  2. Convert to free static Space: browser-side model + voice/text UI

    Verified signature
    17957e0