katostrofik/qwen36-35b-a3b-fp8-two-blackhole-tt-cache
Qwen3.6-35B-A3B-FP8 two-Blackhole TT cache This dataset contains the generated same-source compressed owner-bank cache used by a public Qwen/Qwen3.6-35B-A3B-FP8 two-Blackhole runtime project. Project repo: https://github.com/PMZFX/TT-qwen36-35b-a3b-fp8-two-blackhole The GitHub repo contains the runtime code, TT-Lang spike, reliability harnesses, release notes, and helper scripts. This dataset supplies the generated TT cache that is too large for the GitHub repo.… See the full description on the dataset page: https://huggingface.co/datasets/katostrofik/qwen36-35b-a3b-fp8-two-blackhole-tt-cache.
Qwen3.6-35B-A3B-FP8 two-Blackhole TT cache
This dataset contains the generated same-source compressed owner-bank cache used by a public Qwen/Qwen3.6-35B-A3B-FP8 two-Blackhole runtime project.
Project repo:
https://github.com/PMZFX/TT-qwen36-35b-a3b-fp8-two-blackhole
The GitHub repo contains the runtime code, TT-Lang spike, reliability harnesses, release notes, and helper scripts. This dataset supplies the generated TT cache that is too large for the GitHub repo.
Contents
same_source_compressed_owner_bank_cache_v0/: generated TT tensorbin cache for same-source compressed BFP4 owner-bank residency.same_source_compressed_owner_bank_cache_v0.sha256: SHA-256 checksum manifest for the cache files.
The cache was generated from the warm resident bundle and source tensorbin materialization path in the project. It is intended to let another two-Blackhole system skip the expensive local cache generation step and run the current production candidate directly.
Runtime Profile
Canonical runtime profile:
--mesh-two-card --tt-lang-sidecar-current-production-candidateValidation snapshot from 2026-05-15:
- Public release validation: 8/8 cases passed.
- Generated tokens across validation: 91.
- Token drift mismatches: 0.
- Sidecar fallback markers: 0.
- Sidecar error markers: 0.
- Sidecar setup failures: 0.
- Mean warmed proxy step: about 122 to 129 ms across validation suites.
- Mean warmed routed step: about 14.6 to 15.1 ms across validation suites.
- Observed warmed generation rate: about 7.7 to 8.2 tokens/s in the validation cases.
How to Use
Clone the source repo:
git clone https://github.com/PMZFX/TT-qwen36-35b-a3b-fp8-two-blackhole.git
cd TT-qwen36-35b-a3b-fp8-two-blackholeDownload this dataset into any working directory. If using the GitHub repo's default layout, this example puts it under the repo's results/tt_lang_spike directory:
export PROJECT_ROOT=/path/to/TT-qwen36-35b-a3b-fp8-two-blackhole
hf download katostrofik/qwen36-35b-a3b-fp8-two-blackhole-tt-cache \
--repo-type dataset \
--local-dir "$PROJECT_ROOT/llm/projects/qwen36-35b-a3b-fp8-two-blackhole/results/tt_lang_spike"Verify checksums from the results directory:
cd "$PROJECT_ROOT/llm/projects/qwen36-35b-a3b-fp8-two-blackhole/results/tt_lang_spike"
sha256sum -c same_source_compressed_owner_bank_cache_v0.sha256If the cache is stored elsewhere, pass this manifest path to runtime commands that accept the compressed cache:
--tt-lang-sidecar-compressed-bank-cache-manifest \
/path/to/same_source_compressed_owner_bank_cache_v0/manifest_all_layers_all_roles_cards01_materialized.jsonRun the public release validation from the TT-Lang spike directory:
cd "$PROJECT_ROOT/llm/projects/qwen36-35b-a3b-fp8-two-blackhole/tt_lang_spike"
.venv-tt-lang-py312/bin/python run_public_release_validation.py \
--out-dir ../results/tt_lang_spike/public_release_validation_REVIEWThe source repo also includes fallback scripts to regenerate the cache from local model weights and the warm resident tensorbin bundle when needed.
Notes
This artifact is a generated runtime cache for Tenstorrent Blackhole hardware. It does not replace the upstream model repository or its terms. Use of the cache should follow the upstream Qwen model terms and any applicable project dependencies.
