The story so far
Weights, every dated article, and repo commits on one rail.
22 repo commits since the weights landed
Teal is the weights (2022-05-11), grey ticks are repo commits.
opt-2.7b
facebook Other out 4 years ago
Text in, text out, 2K tokens of context.
Verified 2026-10-25 Rank #548 in LLMs · score 74.9 method v1 Sources 3 Downloads 30d 17.9k · #465 of 619 tracked
Can I use it?OtherCustom license; read the repo LICENSE before commercial hostinglicence file2026-10-24
Do people use it?17.9kdownloads, 30 days#465 of 619 we track · below the median+3 likes, downloads −25% this week94 likesHugging Face2026-10-24
Where it's used
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: 1 quantized builds and 35 Spaces.
- As a service0no API provider listed
- On own hardware1quantized builds
- Built on206 fine-tunes · 11 adapters · 3 merges
- In apps35Hugging Face Spaces
Top Spaces running it: tomg-group-umd/lm-watermarking ♥ 100 · alvanlii/FROMAGe ♥ 27 · SeViLA/SeViLA ♥ 12 · wenkai/FAPM_demo ♥ 11
What people built on it
21 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
21 derivatives on the Hub, most of them adapters
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
Use-case evidence
1 use case
What it is good at 1 use case
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.
Facts
18 fields
Listed in LLMs
Facts
- Context length is 2,048 tokens (2K). (config.json max_position_embeddings, observed 2026-10-24)
- License: Other (friction: review). Custom license; read the repo LICENSE before commercial hosting. (license file, observed 2026-10-24)
- Created 2022-05-11, last modified 2023-09-15. (HF API, observed 2026-10-24)
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | OPTForCausalLM |
| Model type | opt |
| Dtype | float16 |
| MoE | no |
| Context length | 2,048 (config.json max_position_embeddings) |
| Languages | en |
| Papers | arXiv:2205.01068, arXiv:2005.14165 |
| License | Other (other) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · OPTForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · OPTForCausalLM · 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.
All variants
21 variants
Lineage
adapter: 11merge: 3finetune: 6quantized: 1
Top variants (21 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| royokong/prompteol-opt-2.7b | adapter | — | — | — | — | royokong | 0 | 3.9k |
| Optitransfer/Qwen2.5-7B-Instruct-converge-collective-v2 | merge | — | 7.62B | M | other | Optitransfer | 0 | 2k |
| pszemraj/opt-peter-2.7B | finetune | — | 2.65B | M | otherapache-2.0 | pszemraj | 2 | 1.5k |
| Optitransfer/Qwen2.5-7B-Instruct-converge-collective-v1 | merge | — | 7.62B | M | other | Optitransfer | 0 | 1.3k |
| iproskurina/opt-2.7b-GPTQ-4bit-g128 | quantized | GPTQ | 601M | S | other | iproskurina | 0 | 711 |
| Optitransfer/Qwen2.5-7B-Instruct-converge-v1remerge-maskn_project | merge | — | 7.62B | M | other | Optitransfer | 0 | 663 |
| MiniLLM/MiniLLM-OPT-2.7B | finetune | — | — | — | apache-2.0 | MiniLLM | 0 | 606 |
| mahwizzzz/UrduOPT-2.7b-lora | adapter | — | — | — | mit | mahwizzzz | 0 | 49 |
| sert121/orpo_opt2_hh_full | adapter | — | — | — | other | sert121 | 0 | 47 |
| RuoxiL/style-dailymed-from-facebook | adapter | — | — | — | other | RuoxiL | 0 | 46 |
| sert121/orpo_opt2_hh_filtered | finetune | — | 2.65B | M | other | sert121 | 0 | 43 |
| RuoxiL/style-irs-from-facebook-2.7 | adapter | — | — | — | other | RuoxiL | 0 | 37 |
| mynuddin/chatbot | adapter | — | — | — | — | mynuddin | 0 | 37 |
| sert121/orpo_opt2_hh_helpful | adapter | — | — | — | other | sert121 | 0 | 35 |
| VietKien/opt-2.7b-lora-quotes | adapter | — | — | — | other | VietKien | 0 | 30 |
| RuoxiL/style-treasury-from-facebook | adapter | — | — | — | other | RuoxiL | 0 | 29 |
| mrityunjayjha/lora-blip2-vqa-model | adapter | — | — | — | — | mrityunjayjha | 0 | 26 |
| Mashina01/telegram-trained-opt | adapter | — | 2.65B | M | — | Mashina01 | 0 | 24 |
| PoHao/opt2.7-fc1-predictor | finetune | — | — | — | apache-2.0 | PoHao | 0 | 0 |
| PoHao/opt2.7-predictor-guided-sft | finetune | — | — | — | apache-2.0 | PoHao | 0 | 0 |
| PoHao/opt2.7-standard-sft | finetune | — | — | — | apache-2.0 | PoHao | 0 | 0 |
Repository changelog
22 commits
Repository changelog 22 commits
| Date | Commit | SHA |
|---|---|---|
| 2023-09-15 | Update README.md (#25) | 905a4b6 |
| 2023-01-24 | Adding generation config file(s) | 397f71a |
| 2022-06-22 | Update generations after major fix: https://github.com/huggingface/transformers/commit/abc400b06a8ab26cd438b6e9add3aad082ffc48f (#4) | c9c1510 |
| 2022-06-21 | correct checkpoints see: https://github.com/facebookresearch/metaseq/pull/164 | a5227ab |
| 2022-06-16 | Update TF weights (#2) | 4153458 |
| 2022-06-03 | Add flax_weights | 5ef2ffb |
| 2022-06-03 | Add tf_weights | 8bd2e9c |
| 2022-06-03 | Add tf_weights | 0bff200 |
| 2022-05-26 | Change BOS token from 0 to 2 as BOS token is equal to EOS for OPT. See: https://github.com/huggingface/transformers/issues/17431 (#1) | a771dd5 |
| 2022-05-25 | Update tokenizer_config.json | 6c6769e |
| 2022-05-16 | Set commercial attribute to False | 1e18d8f |
| 2022-05-16 | Update README | b24209b |
| 2022-05-13 | Temporarily deactivate the inference API | 90191ab |
| 2022-05-13 | Update config.json | 3810514 |
| 2022-05-12 | Update README.md | 0d79cf9 |
| 2022-05-12 | Update README.md | 7d49491 |
| 2022-05-12 | Update README.md | 4043af6 |
| 2022-05-12 | Create README.md | 1bd7175 |
| 2022-05-12 | finish | 48d8c05 |
| 2022-05-12 | Update config.json | 32872c2 |
| 2022-05-11 | upload 2.7b | 29415cf |
| 2022-05-11 | initial commit | 8ef56f6 |
Papers
2 papers
Papers 2
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).
- Tom B. Brown, Benjamin Mann, Nick Ryder +28 · 2020-05-28 · neural information processing systems · PDF · HF paper page (21 upvotes) · evidence
Similar models
3 models
Similar models
- gpt-j-6b same task (text-generation), permissive license
- DialoGPT-medium same task (text-generation), permissive license
- Qwen-VL-Chat same task (text-generation)
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:42 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:13 UTC · Last verified 2026-10-25
Follow opt-2.7b
Watch this model
- /feeds/models/facebook--opt-2.7b.xml(RSS 2.0) · every change to opt-2.7b
- /feeds/families/facebook--opt.xml(RSS 2.0) · every change across the opt family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.