Up to date
tiiuae Apache 2.0 out 3 years ago
41.8B parameters, text in, text out.
Verified 2026-10-25 Rank #360 in LLMs · score 80.9 method v1 Sources 3 Downloads 30d 7.7k · #523 of 619 tracked
Picked by rule from 19 tracked pieces: the lab's own post, the most shared, a field report, the latest.
On 13 May, 2024, the Technology Innovation Institute released Falcon 2, an open-source model series where Falcon 2 11B scores 64.28 on the Hugging Face leaderboard, narrowly trailing Google's Gemma 7B at 64.29.
The 3 picks above, plus these 16. 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 100 Spaces.
Top Spaces running it: pliny-the-prompter/obliteratus ♥ 621 · HuggingFaceH4/falcon-chat ♥ 557 · h2oai/h2ogpt-chatbot ♥ 295 · h2oai/h2ogpt-chatbot2 ♥ 167
Weights only, an estimate: leave room for the KV cache. Method
About 25 GB at INT4, 50 GB at FP8, 100 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 (2023-05-24), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
21 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.
Falcon-40B is a 41.8B parameter causal decoder-only model designed for text generation.[1][2] The model was trained on 1,000B tokens of RefinedWeb and is released under the Apache 2.0 license.[3][4]
Generated 2026-10-07 15:09 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Listed in LLMs
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | FalconForCausalLM |
| Model type | falcon |
| Dtype | bfloat16 |
| Params by dtype | BF16: 41,835,970,560 |
| Size band | XL |
| MoE | no |
| Languages | en, de, es, fr |
| Papers | arXiv:2205.14135, arXiv:1911.02150, arXiv:2101.00027, arXiv:2005.14165, arXiv:2104.09864, arXiv:2306.01116 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · FalconForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · FalconForCausalLM · 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 falcon-40b: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 100 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 50 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 25 GB · Single high-end GPU (<80 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 4finetune: 5adapter: 12
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| mradermacher/falcon-40b-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 7k |
| Prannesshkva/Ael-Pro-40B | finetune | — | 41.8B | XL | cc-by-nc-nd-4.0 | Prannesshkva | 3 | 5.3k |
| tensorblock/falcon-40b-GGUF | quantized | GGUF | — | — | apache-2.0 | tensorblock | 0 | 3.9k |
| dfurman/Falcon-40B-Chat-v0.1 | adapter | — | — | — | apache-2.0 | dfurman | 38 | 1.4k |
| mradermacher/falcon-40b-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 3.1k |
| BramVanroy/falcon-40b-ft-alpaca-dolly-dutch | finetune | — | 41.3B | XL | cc-by-nc-4.0 | BramVanroy | 4 | 865 |
| Sandiago21/falcon-40b-prompt-answering | quantized | — | — | — | other | Sandiago21 | 1 | 502 |
| nmitchko/medfalcon-v2-40b-lora | adapter | — | — | — | cc-by-nc-4.0 | nmitchko | 3 | 214 |
| monsterapi/Falcon_40B_dolly15k | adapter | — | — | — | apache-2.0 | monsterapi | 0 | 147 |
| Defalt-404/LoRA_Falcon | adapter | — | — | — | — | Defalt-404 | 0 | 134 |
| oscorrea/scores-falcon40b-sm | adapter | — | — | — | — | oscorrea | 0 | 81 |
| oof-baroomf/MuseGPT | adapter | — | — | — | — | oof-baroomf | 0 | 68 |
| sarath7974/Falcon-40B-16bit_finetuned | adapter | — | — | — | apache-2.0 | sarath7974 | 0 | 47 |
| TopperThijs/Falcon-40b-complete-10-15-longer-time | adapter | — | — | — | apache-2.0 | TopperThijs | 0 | 36 |
| zrx-kishore/falcon-40b-4bit | adapter | -4bit | — | — | — | zrx-kishore | 0 | 35 |
| WouterMK/QASD | adapter | — | — | — | — | WouterMK | 0 | 33 |
| TopperThijs/Falcon-40b-complete-10-15 | adapter | — | — | — | apache-2.0 | TopperThijs | 0 | 25 |
| maidacundo/falcon_40b_qlora_sql_r64 | finetune | — | — | — | apache-2.0 | maidacundo | 0 | 0 |
| Acadys/PointCon-Falcon40B | finetune | — | — | — | apache-2.0 | Acadys | 0 | 0 |
| regisss/falcon-40b-lora | finetune | — | — | — | apache-2.0 | regisss | 0 | 0 |
| Jagad1234unique/12 | adapter | — | — | — | — | Jagad1234unique | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2024-08-09 | Adding `safetensors` variant of this model (#112) | 05ab2ee |
| 2023-09-29 | Move to in-library checkpoint (for real this time) (#107) | 4a70170 |
| 2023-09-05 | Clean-up README.md (call for proposal has ended) | 3d7c590 |
| 2023-07-13 | Revert in-library PR (#90) | 561820f |
| 2023-07-12 | Move to in-library checkpoint (#81) | f1ba7d3 |
| 2023-06-15 | Update README.md | c47b371 |
| 2023-06-09 | Update README.md (#40) | 2ac60b0 |
| 2023-06-09 | Add usage recommendations | 2e3a7d3 |
| 2023-06-05 | Update citation info | 88b92ae |
| 2023-05-31 | Remove TII Falcon LLM license | b046281 |
| 2023-05-31 | Update license info to Apache 2.0 | fdc4d06 |
| 2023-05-31 | `license: apache-2.0` (#28) | 337bddb |
| 2023-05-31 | Enable loading with auto classes (#23) | f431193 |
| 2023-05-30 | license: tii-falcon-llm (#19) | eddb07e |
| 2023-05-30 | Update config.json | 52aee9e |
| 2023-05-30 | Update README.md | 7324a79 |
| 2023-05-30 | corrected a typo (#5) | c63a64a |
| 2023-05-30 | Update modelling_RW.py | e49c179 |
| 2023-05-26 | Update model card for public release | 6e61c89 |
| 2023-05-26 | Add license file | 31c2141 |
| 2023-05-25 | Update README.md | 5887b76 |
| 2023-05-25 | Update modelling_RW.py | 7c4c297 |
| 2023-05-25 | Create README.md | ebc2a3c |
| 2023-05-25 | Upload RWForCausalLM | 56dde4b |
| 2023-05-24 | Upload tokenizer | 6064d60 |
| 2023-05-24 | Upload RWForCausalLM | e7950c4 |
| 2023-05-24 | initial commit | faeeeb3 |
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 21:46 UTC · Summary: 2026-10-07 15:09 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.
