Up to date
bigcode BigCode OpenRAIL-M out 3 years ago
3.03B parameters, text in, text out, 16K tokens of context.
Verified 2026-10-25 Rank #475 in LLMs · score 77.2 method v1 Sources 3 Downloads 30d 29.9k · #423 of 619 tracked
Picked by rule from 9 tracked pieces: the lab's own post, the most shared, a field report, the latest.
SelfCodeAlign creates StarCoder2-Instruct by using StarCoder2-15B to generate thousands of instruction-response pairs for fine-tuning StarCoder-15B without human annotations or distilled data from proprietary models.
StarCoder2-15B-Instruct achieves a 72.6 HumanEval score, surpassing the 72.0 score of CodeLlama-70B-Instruct.
The 2 picks above, plus these 7. 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: 32 quantized builds and 89 Spaces.
Top Spaces running it: KBaba7/Quant ♥ 51 · FallnAI/Quantize-HF-Models ♥ 19 · bhaskartripathi/LLM_Quantization ♥ 16 · openfree/LLM_Quantization ♥ 16
Weights only, an estimate: leave room for the KV cache. Method
About 1.8 GB at INT4, 3.6 GB at FP8, 7.3 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 (2023-11-29), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.
375 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.
The architecture utilizes grouped-query and sliding window attention with a context window of 16,384 tokens.[1]
Generated 2026-10-07 15:11 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
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 | Starcoder2ForCausalLM |
| Model type | starcoder2 |
| Params by dtype | F32: 3,030,371,328 |
| Size band | M |
| MoE | no |
| Context length | 16,384 (config.json max_position_embeddings) |
| Papers | arXiv:2305.13245, arXiv:2205.14135, arXiv:2004.05150, arXiv:2207.14255, arXiv:2402.19173 |
| License | BigCode OpenRAIL-M (bigcode-openrail-m) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Starcoder2ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Starcoder2ForCausalLM · 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 starcoder2-3b: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 7.3 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 3.6 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 1.8 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 32adapter: 291finetune: 51merge: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| second-state/StarCoder2-3B-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | second-state | 20 | 59k |
| mradermacher/starcoder2-3b-i1-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | mradermacher | 0 | 17.8k |
| QuantFactory/starcoder2-3b-instruct-v0.1-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | QuantFactory | 1 | 10.6k |
| mradermacher/starcoder2-3b-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | mradermacher | 3 | 8.6k |
| DevQuasar-9/bigcode.starcoder2-3b-GGUF | quantized | GGUF | — | — | — | DevQuasar-9 | 1 | 7.4k |
| tensorblock/starcoder2-3b-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | tensorblock | 2 | 4k |
| smangrul/starcoder-3b-hugcoder | adapter | — | — | — | bigcode-openrail-m | smangrul | 0 | 3.8k |
| Ffftdtd5dtft/starcoder2-3b-Q2_K-GGUF | quantized | Q2_K | — | — | bigcode-openrail-m | Ffftdtd5dtft | 0 | 1.9k |
| ysn-rfd/starcoder2-3b-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | ysn-rfd | 1 | 1.7k |
| combos/starcoder2-3b-Q4_K_M-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | combos | 1 | 1.2k |
| onekq-ai/starcoder2-3b-bnb-4bit | quantized | -bnb | 3.12B | M | bigcode-openrail-m | onekq-ai | 0 | 550 |
| Sayan18/finetune_starcoder2 | adapter | — | — | — | bigcode-openrail-m | Sayan18 | 3 | 349 |
| raghav0/starcoder2-3b-Q4_K_M-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | raghav0 | 0 | 482 |
| MuratKomurcu/starcoder2-stm32 | finetune | — | 3.03B | M | mit | MuratKomurcu | 2 | 377 |
| OmAr-Kader-DeV/starcoder2-3b-mlx-8Bit | quantized | mlx | 3.03B | M | bigcode-openrail-m | OmAr-Kader-DeV | 0 | 436 |
| SKT-NRS/STYLE-CODER-1 | finetune | — | 3.03B | M | bigcode-openrail-m | SKT-NRS | 2 | 272 |
| onekq-ai/starcoder2-3b-instruct-v0.1 | finetune | — | 3.03B | M | bigcode-openrail-m | onekq-ai | 0 | 368 |
| alal123/starcoder2-3b-Q4_K_M-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | alal123 | 0 | 356 |
| joehiggi/gcode-specialist-3b | adapter | GGUF | — | — | bigcode-openrail-m | joehiggi | 0 | 350 |
| xfel/starcoder2-3b-Q4_K_M-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | xfel | 0 | 327 |
| vwdvaan/starcoder2-3b-Q4_K_M-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | vwdvaan | 0 | 317 |
| MuratKomurcu/starcoder2-stm32-turkish | finetune | — | — | — | bigcode-openrail-m | MuratKomurcu | 6 | 0 |
| oakela/starcoder2_dev_data | finetune | — | 3.03B | M | bigcode-openrail-m | oakela | 0 | 296 |
| ekryski/starcoder2-3b-4bit | quantized | -4bit | 3.03B | M | apache-2.0 | ekryski | 0 | 270 |
| ibkya/Arden-World-3B | quantized | GGUF | 3.03B | M | apache-2.0 | ibkya | 0 | 265 |
| ShayonSarker/StarCoder2-3B-GGUF | quantized | GGUF | — | — | bigcode-openrail-m | ShayonSarker | 1 | 204 |
| zhenchonghu/starcoder2-3b-alpaca-lora | adapter | — | — | — | — | zhenchonghu | 0 | 247 |
| RohamKoohestani/finetune_starcoder2_3b_tokens_added_6k_steps | finetune | — | 3.03B | M | — | RohamKoohestani | 0 | 241 |
| gubee/fine-tuned-model | quantized | GGUF | 3.03B | M | bigcode-openrail-m | gubee | 0 | 236 |
| arthur-chien-0530/sft_only | finetune | — | 3.03B | M | — | arthur-chien-0530 | 0 | 234 |
| CodeHima/finetunedPHP_starcoder2 | adapter | — | — | — | bigcode-openrail-m | CodeHima | 0 | 212 |
| RohamKoohestani/finetune_starcoder2_3b_tokens_added | finetune | — | 3.03B | M | — | RohamKoohestani | 0 | 204 |
| prasannadhungana8848/finetune_starcoder2_with_R_data | adapter | — | — | — | bigcode-openrail-m | prasannadhungana8848 | 0 | 192 |
| Spandan98/SAS_Metadata_Adaptor | adapter | — | — | — | — | Spandan98 | 0 | 159 |
| smangrul/starcoder-3b-hugcoder-loftq | adapter | — | — | — | bigcode-openrail-m | smangrul | 0 | 157 |
| andstor/bigcode-starcoder2-3b-unit-test-fine-tuning | finetune | — | 3.03B | M | bigcode-openrail-m | andstor | 0 | 128 |
| SenalVithana/gesture-to-llm-adapter | finetune | — | 3.03B | M | mit | SenalVithana | 0 | 126 |
| oakela/finetune_starcoder2_nate_continue_600steps | finetune | — | 3.03B | M | bigcode-openrail-m | oakela | 0 | 125 |
| oakela/finetune_starcoder2_nolora | finetune | — | 3.03B | M | bigcode-openrail-m | oakela | 0 | 119 |
| oakela/finetune_starcoder2_cleaned | finetune | — | 3.03B | M | bigcode-openrail-m | oakela | 0 | 116 |
| Date | Commit | SHA |
|---|---|---|
| 2024-03-04 | add dropout (#6) | 733247c |
| 2024-03-01 | fix paper (#3) | cc9b0d4 |
| 2024-02-29 | Add architectures field to config.json (#2) | 54da767 |
| 2024-02-28 | Update README.md | 444cd6f |
| 2024-02-27 | Update README.md (#1) | 4e1a8cf |
| 2024-02-27 | Update README.md | 421d645 |
| 2024-02-27 | rewrite | 73551f5 |
| 2024-02-27 | Update README.md | 107effb |
| 2024-02-27 | Update README.md | 2cdb3d3 |
| 2024-02-26 | Create README.md | 088e2ed |
| 2024-02-26 | fix eos token | 8f2ca22 |
| 2024-02-26 | add missing tokenizer files | 6af7c4b |
| 2024-02-23 | safetensors | 56b73d9 |
| 2024-02-22 | remove obsolete info in readme | 9eebcd0 |
| 2024-02-22 | update config and model (new modeling) | f14327e |
| 2024-01-15 | Update README.md | 11a6bb4 |
| 2024-01-15 | Create README.md | d65d24c |
| 2023-11-29 | Upload tokenizer.json | b75eb95 |
| 2023-11-29 | 50000 | 9a1f939 |
| 2023-11-29 | initial commit | 09a9c4c |
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 15:11 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.
