Up to date
EleutherAI Apache 2.0 out 4 years ago
213M parameters, text in, text out, 2K tokens of context.
Verified 2026-10-25 Rank #254 in LLMs · score 85.7 method v1 Sources 3 Downloads 30d 3.4M · #44 of 619 tracked
Picked by rule from 15 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Pythia is a suite of 16 large language models all trained on public data seen in the exact same order and ranging in size from 70M to 12B parameters.
The 4 picks above, plus these 11. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 30 quantized builds and 100 Spaces.
Top Spaces running it: pliny-the-prompter/obliteratus ♥ 621 · KBaba7/Quant ♥ 51 · EvanTHU/MotionLLM ♥ 43 · FallnAI/Quantize-HF-Models ♥ 19
Weights only, an estimate: leave room for the KV cache. Method
About 0.1 GB at INT4, 0.3 GB at FP8, 0.5 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2023-02-08), amber triangles are 4 dated articles (each links to it), grey ticks are repo commits.
586 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in LLMs
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | GPTNeoXForCausalLM |
| Model type | gpt_neox |
| Dtype | float16 |
| Params by dtype | U8: 50,331,648 · F16: 162,323,040 |
| Size band | S |
| MoE | no |
| Context length | 2,048 (config.json max_position_embeddings) |
| Languages | en |
| Papers | arXiv:2304.01373, arXiv:2101.00027, arXiv:2201.07311 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · GPTNeoXForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for pythia-160m: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 0.5 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.3 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.1 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 30finetune: 380adapter: 176
| Date | Commit | SHA |
|---|---|---|
| 2023-07-09 | Adding `safetensors` variant of this model (#2) | 50f5173 |
| 2023-06-08 | fix batch size and add paper | 26b94a3 |
| 2023-04-03 | Upload pytorch_model.bin | 57c4cbf |
| 2023-04-03 | Upload tokenizer.json | 40f6318 |
| 2023-04-03 | Upload tokenizer_config.json | 4560d92 |
| 2023-04-03 | Upload config.json | 521f765 |
| 2023-04-03 | Upload special_tokens_map.json | 7e30bb8 |
| 2023-03-29 | create updated readme | 64a5b91 |
| 2023-03-23 | Upload config.json | e0a0251 |
| 2023-03-23 | Upload special_tokens_map.json | 5c39738 |
| 2023-03-23 | Upload tokenizer.json | 4c6801b |
| 2023-03-23 | Upload pytorch_model.bin | 0243faf |
| 2023-03-23 | Upload tokenizer_config.json | a2e7dc4 |
| 2023-03-22 | Upload config.json | 610be53 |
| 2023-03-22 | Upload pytorch_model.bin | ac17eb6 |
| 2023-03-22 | Upload special_tokens_map.json | a6b1312 |
| 2023-03-22 | Upload tokenizer_config.json | 30ac103 |
| 2023-03-22 | Upload tokenizer.json | e1fd0cf |
| 2023-03-22 | Upload special_tokens_map.json | 1dc1688 |
| 2023-03-22 | Upload config.json | cd86582 |
| 2023-03-22 | Upload tokenizer_config.json | a82aaf0 |
| 2023-03-22 | Upload tokenizer.json | d38718f |
| 2023-03-22 | Upload pytorch_model.bin | 4fb48c3 |
| 2023-03-22 | Upload config.json | 6e10859 |
| 2023-03-22 | Upload tokenizer_config.json | 5e3f0aa |
| 2023-03-22 | Upload pytorch_model.bin | e4d63e7 |
| 2023-03-22 | Upload special_tokens_map.json | 4550161 |
| 2023-03-22 | Upload tokenizer.json | f96ca98 |
| 2023-03-22 | Upload config.json | 7acef1e |
| 2023-03-22 | Upload special_tokens_map.json | 2be6e37 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:38 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:08 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
