← all publishers

modelcloud

@modelcloud source repo

3 published skills

  1. Forward Kernel Validation · modelcloud
    Validate optimized Recirculation forward kernels against an unfused, unbatched oracle when changing batching, fusion, attention, CUDA graphs, or accelerator inference code.
    0
    installs
  2. Gptqmodel Lazy Load Smoke Testing · modelcloud
    End-to-end smoke test for GPT-QModel's lazy checkpoint load (LazyTurtle) and first-layer materialization through StageInputsCapture.cache_inputs. Use when verifying lazy-load behavior, first-layer input capture, shell_module_materialize module_path propagation, or CPU-only GPTQ quantization from a tiny fixture.
    0
    installs
  3. Gptqmodel Tokenizer Normalization · modelcloud bundle
    Diagnose and correct GPT-QModel tokenizer initialization, tokenization normalization, special-token handling, prompt rendering, and chat-template problems. Use when inference or evaluation quality suggests wrong token IDs or prompts, when a model needs tokenizer load kwargs or compatibility patches, or when tokenizer behavior is being changed in GPT-QModel; reusable corrective behavior must be implemented, tested, versioned, pushed, and submitted upstream to github.com/ModelCloud/Tokenicer.
    0
    installs