category stringclasses 6
values | reason stringclasses 6
values | matched_pattern stringlengths 6 40 | model_id stringlengths 18 77 | downloads int64 17 12.2M | torch_tpu_version stringclasses 3
values | sweep_id stringclasses 3
values |
|---|---|---|---|---|---|---|
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=google/gemma-4-31B-it | farbodtavakkoli/OTel-2.0-LLM-31B-IT | 7,084,920 | 0.1.1.dev20260825050227 | 20260825-143051 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | orcarouter/Qwen3.8-27B-Uncensored-MLX | 68,855 | 0.1.1.dev20260825050227 | 20260825-143051 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | abliterated | huihui-ai/Huihui-Qwen3.8-27B-abliterated | 31,767 | 0.1.1.dev20260825050227 | 20260825-143051 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 | 15,828 | 0.1.1.dev20260825050227 | 20260825-143051 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | orcarouter/Qwen3.8-27B-Uncensored | 15,341 | 0.1.1.dev20260825050227 | 20260825-143051 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=Qwen/Qwen3.5-9B | empero-ai/Qwen3.8-9B-Distill | 13,761 | 0.1.1.dev20260825050227 | 20260825-143051 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=Qwen/Qwen3.5-35B-A3B | apodex/Apodex-1.1-mini | 484 | 0.1.1.dev20260825050227 | 20260825-143051 |
diffusion | Diffusion LM — non-autoregressive inference path | DFlash | incoai/Qwen3.8-27B-DFlash2 | 105,786 | 0.1.1.dev20260825050227 | 20260825-143051 |
diffusion | Diffusion LM — non-autoregressive inference path | DFlash | z-lab/Qwen3.8-27B-DFlash2 | 64,984 | 0.1.1.dev20260825050227 | 20260825-143051 |
non-LLM-modality | OCR / VLA / segmentation / audio modality — not a text LM | whisper | superwhisper/s1-mini | 3,474 | 0.1.1.dev20260825050227 | 20260825-143051 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | -NVFP4 | nvidia/Qwen3.6-35B-A3B-NVFP4 | 12,163,610 | 0.1.1.dev20260825050227 | 20260825-143051 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | quantization_config in hub config | openai/gpt-oss-20b | 7,056,476 | 0.1.1.dev20260825050227 | 20260825-143051 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | quantization_config in hub config | EschaLabs/Qwen3.8-27B-Escha-W2 | 2,319 | 0.1.1.dev20260825050227 | 20260825-143051 |
trust-remote-code | Custom modeling code via auto_map — not part of the stable AutoClass surface | auto_map present | inclusionAI/Ling-3.0-tiny | 16,397 | 0.1.1.dev20260825050227 | 20260825-143051 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=google/gemma-4-31B-it | farbodtavakkoli/OTel-2.0-LLM-31B-IT | 7,017,534 | 0.1.1.dev20260826100239 | 20260826-152322 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | orcarouter/Qwen3.8-27B-Uncensored-MLX | 79,395 | 0.1.1.dev20260826100239 | 20260826-152322 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | abliterated | huihui-ai/Huihui-Qwen3.8-27B-abliterated | 37,177 | 0.1.1.dev20260826100239 | 20260826-152322 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | orcarouter/Qwen3.8-27B-Uncensored | 17,567 | 0.1.1.dev20260826100239 | 20260826-152322 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 | 16,327 | 0.1.1.dev20260826100239 | 20260826-152322 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=Qwen/Qwen3.5-35B-A3B | apodex/Apodex-1.1-mini | 840 | 0.1.1.dev20260826100239 | 20260826-152322 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=tri-fair-lab/Snowdon1.1-Small | thomsonreuters/Thomson-1.0-Small | 214 | 0.1.1.dev20260826100239 | 20260826-152322 |
diffusion | Diffusion LM — non-autoregressive inference path | DFlash | incoai/Qwen3.8-27B-DFlash2 | 124,420 | 0.1.1.dev20260826100239 | 20260826-152322 |
diffusion | Diffusion LM — non-autoregressive inference path | DFlash | z-lab/Qwen3.8-27B-DFlash2 | 87,495 | 0.1.1.dev20260826100239 | 20260826-152322 |
non-LLM-modality | OCR / VLA / segmentation / audio modality — not a text LM | whisper | superwhisper/s1-mini | 3,921 | 0.1.1.dev20260826100239 | 20260826-152322 |
non-generative | Non-generative model class (embedding / reranker / guard head) | Embedding- | Qwen/Qwen3-Embedding-0.6B | 7,022,748 | 0.1.1.dev20260826100239 | 20260826-152322 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | -NVFP4 | nvidia/Qwen3.6-35B-A3B-NVFP4 | 12,017,108 | 0.1.1.dev20260826100239 | 20260826-152322 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | quantization_config in hub config | EschaLabs/Qwen3.8-27B-Escha-W2 | 2,481 | 0.1.1.dev20260826100239 | 20260826-152322 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=google/gemma-4-31b-it | farbodtavakkoli/OTel-2.0-LLM-31B-IT | 6,777,482 | 0.1.1.dev20260902101038 | 20260902-134049 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | orcarouter/Qwen3.8-27B-Uncensored-MLX | 121,028 | 0.1.1.dev20260902101038 | 20260902-134049 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=Qwen/Qwen3.5-35B-A3B | apodex/Apodex-1.1-mini | 2,500 | 0.1.1.dev20260902101038 | 20260902-134049 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | base_model=tri-fair-lab/Snowdon1.1-Small | thomsonreuters/Thomson-1.0-Small | 1,130 | 0.1.1.dev20260902101038 | 20260902-134049 |
derivative-fork | Finetune / post-training fork of a base model (abliterated / uncensored / heretic / derestricted / distilled). Shares its base's architecture and contributes no new signal, so we sweep the base instead. | uncensored | DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU | 17 | 0.1.1.dev20260902101038 | 20260902-134049 |
diffusion | Diffusion LM — non-autoregressive inference path | DFlash | incoai/GLM-5.3-Flash-DFlash2 | 9,190 | 0.1.1.dev20260902101038 | 20260902-134049 |
non-LLM-modality | OCR / VLA / segmentation / audio modality — not a text LM | whisper | superwhisper/s1-mini | 6,970 | 0.1.1.dev20260902101038 | 20260902-134049 |
non-generative | Non-generative model class (embedding / reranker / guard head) | Embedding- | Qwen/Qwen3-Embedding-0.6B | 6,778,015 | 0.1.1.dev20260902101038 | 20260902-134049 |
non-generative | Non-generative model class (embedding / reranker / guard head) | Embedding- | tencent/WeMM-Embedding-9B | 2,618 | 0.1.1.dev20260902101038 | 20260902-134049 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | -NVFP4 | nvidia/Qwen3.6-35B-A3B-NVFP4 | 10,606,938 | 0.1.1.dev20260902101038 | 20260902-134049 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | quantization_config in hub config | openai/gpt-oss-20b | 6,519,052 | 0.1.1.dev20260902101038 | 20260902-134049 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | -NVFP4 | RadixArk/Qwen3.8-Flash-Next-NVFP4 | 133,491 | 0.1.1.dev20260902101038 | 20260902-134049 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | -NVFP4 | LibertAIDAI/GLM-5.3-Flash-NVFP4 | 24,997 | 0.1.1.dev20260902101038 | 20260902-134049 |
quantized | Quantized model — TPU has no loader for AWQ / GPTQ / compressed-tensors / bnb | -NVFP4 | QUASAR-QAT/Qwen3.8-27B-QUASAR-NVFP4 | 6,099 | 0.1.1.dev20260902101038 | 20260902-134049 |
trust-remote-code | Custom modeling code via auto_map — not part of the stable AutoClass surface | auto_map present | pipecat-ai/phonellm-alpha-1 | 6,813 | 0.1.1.dev20260902101038 | 20260902-134049 |
trust-remote-code | Custom modeling code via auto_map — not part of the stable AutoClass surface | auto_map present | XHToken/Spark-X2.5-4B | 429 | 0.1.1.dev20260902101038 | 20260902-134049 |
torch-tpu-model-support
Support sweep results for HuggingFace models on Google Cloud TPU with torch_tpu.
Each row is one model tested in one sweep. The sweep loads the model and
generates 8 tokens with the fixed prompt "The future of AI is" in bf16.
Result values
PASS— model loaded and generated successfullyFAIL(load)— exception during model loadFAIL(gen)— load succeeded, generation raisedFAIL(output)— generation returned but the text was judged not valid EnglishFAIL(timeout)— killed after exceeding the sweep wall-clock budgetFAIL(other)— unclassified failure (render error, OOM, etc.)
Output judge
english— verdict of a small CPU instruct model asked whether the generated text is sensible English continuing the prompt:Y,N, ornullfor runs that predate the judge (or where it could not run).
Token counts
tokens_in— number of tokens submitted as input (after any chat templating for instruction-tuned models)tokens_out— number of tokens generated beyond the input- Both columns are
nullfor runs from sweeps that predate token counting.
Links
- Source repo: https://github.com/huggingface/tpu-test
- Raw logs bucket:
hf://buckets/{user}/torch-tpu-transformers/
- Downloads last month
- 103