Up to date
Qwen tongyi-qianwen-license-agreement out 3 years ago
7.72B parameters, text in, text out, 32K tokens of context.
Verified 2026-10-25 Rank #412 in LLMs · score 79.2 method v1 Sources 3 Downloads 30d 68.1k · #361 of 619 tracked
Picked by rule from 36 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Strata runs the 125-billion-parameter Qwen3.8-Flash-Next model on consumer hardware under an MIT license with no cloud dependencies.
The 2 picks above, plus these 23. 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: 4 quantized builds and 31 Spaces.
Top Spaces running it: LanguageBind/MoE-LLaVA ♥ 154 · Justinrune/LLaMA-Factory ♥ 12 · officialhimanshu595/llama-factory ♥ 4
Weights only, an estimate: leave room for the KV cache. Method
About 4.6 GB at INT4, 9.3 GB at FP8, 19 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 (2023-08-03), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
10 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.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported C-Eval · Work across languages | 63.5 | 5-shot | model author | 2026-10-07 |
| author-reported GSM8K · Reason through hard problems | 51.7 | 8-shot | model author | 2026-10-07 |
| author-reported MATH-500 · Reason through hard problems | 11.6 | 4-shot | model author | 2026-10-07 |
| author-reported MMLU · Reason through hard problems | 58.2 | 5-shot | model author | 2026-10-07 |
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 | QWenLMHeadModel |
| Model type | qwen |
| Params by dtype | BF16: 7,721,324,544 |
| Size band | M |
| MoE | no |
| Context length | 32,768 (config.json max_position_embeddings) |
| Languages | zh, en |
| Papers | arXiv:2309.16609 |
| License | tongyi-qianwen-license-agreement (other) |
| Hub snapshot | 2026-10-25 |
vLLM: not in registry · QWenLMHeadModel · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · QWenLMHeadModel · 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 Qwen-7B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 19 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 9.3 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 4.6 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
finetune: 3quantized: 4adapter: 3
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| rinna/nekomata-7b | finetune | — | 7.72B | M | other | rinna | 7 | 23.9k |
| Locutusque/UltraQwen-7B | finetune | — | 7.72B | M | other | Locutusque | 2 | 8.5k |
| mradermacher/Qwen-7B-i1-GGUF | quantized | GGUF | — | — | other | mradermacher | 0 | 6.7k |
| mradermacher/Qwen-7B-GGUF | quantized | GGUF | — | — | other | mradermacher | 0 | 3.1k |
| h31337/Qwen-7B-GGUF | quantized | GGUF | — | — | — | h31337 | 0 | 849 |
| mohit95559/mymodel | finetune | — | 7.72B | M | other | mohit95559 | 0 | 238 |
| QLWD/RepoShiled-7b-AWQ | quantized | AWQ | 7.62B | M | — | QLWD | 0 | 212 |
| quyanh/qwen-7b-neurips-v1 | adapter | — | — | — | — | quyanh | 0 | 89 |
| fromson99/FinMindAI_model | adapter | — | — | — | — | fromson99 | 0 | 19 |
| fengn/llama-factory-test | adapter | — | — | — | other | fengn | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2024-01-04 | Update README.md | ef3c5c9 |
| 2023-12-13 | update wechat | ffe04dd |
| 2023-12-13 | update wechat | ad6d04e |
| 2023-12-07 | update modeling_qwen.py | d4efd21 |
| 2023-12-06 | update modeling_qwen.py | 98d83d3 |
| 2023-12-04 | update modeling_qwen.py | 1ff2086 |
| 2023-12-03 | update modeling_qwen.py | 23a5ae1 |
| 2023-11-30 | update | e7a368b |
| 2023-11-14 | remove fix-sized causal mask | f7bc352 |
| 2023-11-14 | update wechat.png | d135dce |
| 2023-11-05 | add kernel file check in modeling_qwen.py | 7b08ae1 |
| 2023-10-26 | update modeling.py | 6b28d40 |
| 2023-10-16 | update modeling_qwen.py | cfe0ba3 |
| 2023-10-14 | update batch inference | e08ed08 |
| 2023-10-11 | format | c9bdb95 |
| 2023-10-09 | update tokenization.py | e7abfa5 |
| 2023-10-09 | update tokenization.py | 39e911d |
| 2023-10-08 | Update README.md (#14) | 3484186 |
| 2023-09-29 | Update README.md | d3f8e6a |
| 2023-09-28 | softmax_in_fp32 | a6ca629 |
| 2023-09-27 | update modeling_qwen.py | 6837ccb |
| 2023-09-27 | update kernels | dd02cde |
| 2023-09-26 | update modeling_qwen.py | 9078774 |
| 2023-09-25 | update modeling_qwen.py | ea57eef |
| 2023-09-25 | update kvcache | b514838 |
| 2023-09-25 | update scores | 17d0047 |
| 2023-09-24 | update readme | 278ca1a |
| 2023-09-24 | Upload tokenizer.png | 9284a07 |
| 2023-09-24 | update batch infer | cf1a074 |
| 2023-09-24 | update model | abcb6d6 |
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 21:39 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:10 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.
