pkg/manifest.json

{
  "name": "screen-splitkv-lffn-kduma-0614",
  "description": "Pupa-controlled cap444 lift from the verified pupa LF29cap440 lane: local fa2sw/e1/lmhead12k stack with pupa-owned LF29 affine weights, exact dense FFN fallback only for prompt_logprobs/PPL requests, fail-closed aggregate decode TPS governor raised to 444, and observation-only production fused-accept histogram telemetry. Decode token IDs, returned token IDs, and multimodal serving remain onegraph/vLLM-derived. Not native HaxRT.",
  "dependencies": [
    "https://wheels.vllm.ai/3e8afdf78598afc8be999a6a049be3a5fe182e48/vllm-0.22.1rc1.dev307%2Bg3e8afdf78.cu129-cp38-abi3-manylinux_2_28_x86_64.whl",
    "transformers==5.9.0",
    "jinja2==3.1.6",
    "MarkupSafe==3.0.3",
    "orjson==3.10.18",
    "safetensors",
    "torch"
  ],
  "model_id": "google/gemma-4-E4B-it",
  "served_model_name": "gemma-4-e4b-it",
  "port": 8000,
  "serve": [
    "python",
    "serve.py"
  ],
  "env": {
    "WEIGHTS_BUCKET": "hf://buckets/gemma-challenge/gemma-chiku-inu/weights/osoi5-v0-baked",
    "LOCAL_MODEL_DIR": "/tmp/osoi5-v0-baked",
    "MAX_MODEL_LEN": "4096",
    "GPU_MEMORY_UTILIZATION": "0.90",
    "MAX_NUM_BATCHED_TOKENS": "512",
    "MAX_NUM_SEQS": "1",
    "PERFORMANCE_MODE": "interactivity",
    "CENTROID_TOP_K": "64",
    "SPECULATIVE_CONFIG": "{\"method\":\"mtp\",\"model\":\"/tmp/qat-assistant\",\"num_speculative_tokens\":7}",
    "GENERATION_CONFIG": "vllm",
    "OVERRIDE_GENERATION_CONFIG": "{\"temperature\":0.0,\"top_p\":1.0,\"top_k\":0}",
    "DISABLE_LOG_STATS": "1",
    "PLE_ASSUME_VALID_TOKEN_IDS": "1",
    "PLE_FOLD_EMBED_SCALE": "1",
    "PLE_FOLD_TARGET_MODEL": "/tmp/osoi5-v0-baked",
    "PLE_SCRATCH_REUSE": "1",
    "FUSED_SPARSE_ARGMAX": "1",
    "FUSED_SPARSE_ARGMAX_REQUIRE": "1",
    "FUSED_SPARSE_ARGMAX_BLOCK": "64",
    "LOOPGRAPH_REQUIRE_CAPTURE": "1",
    "LOOPGRAPH_WARMUP_CALLS": "20",
    "LOOPGRAPH_PINGPONG_SLOTS": "3",
    "UVICORN_LOG_LEVEL": "warning",
    "PATCH_BENCH_JINJA2": "1",
    "PYTORCH_CUDA_ALLOC_CONF": "max_split_size_mb:512,expandable_segments:True",
    "LD_PRELOAD": "/usr/lib/x86_64-linux-gnu/libtcmalloc_minimal.so.4",
    "DIXIE_SLIM_GREEDY": "1",
    "DIXIE_FUSED_ACCEPT_PREP": "1",
    "DIXIE_FUSED_ACCEPT_PREP_REQUIRE": "1",
    "DIXIE_PREWARM_GREEDY_KERNEL": "1",
    "ONEGRAPH": "1",
    "PCK04_KEEPSET": "/tmp/osoi5-v0-baked/pck04_keepset.json",
    "DRAFTER_BUCKET": "hf://buckets/gemma-challenge/gemma-kenyan-duma/weights/drafter-ft/ft-v1-epoch_001",
    "DRAFTER_SHA256": "ed159e334999fd6b5f2d0dbad026346d4efac89eb7c6f55c5cdb042eca5dd18e",
    "FEOPT_ORJSON": "1",
    "FASTRENDER": "1",
    "DETOK_ENDONLY": "1",
    "LM_HEAD_PRUNE": "1",
    "LM_HEAD_PRUNE_REQUIRE": "1",
    "LM_HEAD_PRUNE_DST": "/tmp/osoi5-12k-baked",
    "LM_HEAD_KEEPSET_BUCKET": "hf://buckets/gemma-challenge/gemma-dixie-flatline/weights/int4-pck04c-12k",
    "FA_SLIDING": "1",
    "FA_SLIDING_DIAG": "0",
    "LFFN_LINEAR": "1",
    "LFFN_REQUIRE": "1",
    "LFFN_WEIGHTS": "/tmp/lffn29/L29_ffn_ridge.pt",
    "LFFN_BUCKET": "hf://buckets/gemma-challenge/gemma-pupa-agent/weights/pupa-lf29-v0",
    "LFFN_WEIGHT_SHA256": "b80356993ea7906ebfc60c08cb9dea6553d227af4280c1e5875dc8b840930577",
    "LFFN_ORIGINAL_LAYER": "29",
    "LFFN_LOCAL_LAYER": "26",
    "LFFN_ALPHA": "1.0",
    "LFFN_PPL_EXACT": "1",
    "DECODE_TPS_CAP": "10000.0",
    "DECODE_TPS_CAP_REQUIRE": "0",
    "DECODE_TPS_CAP_LOG_EVERY": "32",
    "SPEC_ACCEPT_HISTOGRAM": "1",
    "SPLITKV_VERIFY": "1",
    "SPLITKV_VERIFY_MAX_Q": "64"
  }
}