Up to date
Models › translategemma family
google Gemma Terms of Use out 9 months ago
12.2B parameters, images and text in, text out, 128K tokens of context.
Verified 2026-10-25 Rank #652 in LLMs · score 73.1 method v1 Sources 3 Downloads 30d 8k · #522 of 619 tracked
Changed this week: 2 new quantized builds on the Hub (33 in all).2026-10-24 · 1 new quantized build on the Hub (32 in all).2026-10-24
Picked by rule from 3 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Google introduces TranslateGemma, a suite of open translation models built on Gemma 3 available in 4B, 12B, and 27B parameter sizes.
The 1 pick above, plus these 2. 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: 33 quantized builds and 5 Spaces.
Top Spaces running it: hysts/translategemma-27b-it ♥ 34 · zxc4wewewe/New_folder_2 ♥ 1 · build-small-hackathon/Vernacular ♥ 1
Weights only, an estimate: leave room for the KV cache. Method
About 7.3 GB at INT4, 15 GB at FP8, 29 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 (2026-01-12), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.
46 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 Vistra* (4 langs) MetricX ↓ | 2.57 | 4B | model author | 2026-10-07 |
| author-reported Vistra* (4 langs) MetricX ↓ | 2.08 | 12B | model author | 2026-10-07 |
| author-reported Vistra* (4 langs) MetricX ↓ | 1.57 | 27B | model author | 2026-10-07 |
| author-reported WMT24++ · Work across languages | 5.32 | 4B | model author | 2026-10-07 |
| author-reported WMT24++ · Work across languages | 3.6 | 12B | model author | 2026-10-07 |
| author-reported WMT24++ · Work across languages | 3.09 | 27B | model author | 2026-10-07 |
| author-reported WMT25 (10 langs) MQM ↓ | 7.94 | 12B | model author | 2026-10-07 |
| author-reported WMT25 (10 langs) MQM ↓ | 5.86 | 27B | model author | 2026-10-07 |
TranslateGemma is a 13.2B parameter image-text-to-text model based on the Gemma 3 family.[1][2] The model performs translation tasks across 55 languages using text or image inputs.[3]
Generated 2026-10-07 15:12 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 | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Gemma3ForConditionalGeneration |
| Model type | gemma3 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 12,187,325,040 |
| Size band | L |
| MoE | no |
| Context length | 131,072 (config.json max_position_embeddings) |
| Gated | manual |
| Papers | arXiv:2601.09012, arXiv:2503.19786 |
| License | Gemma Terms of Use (gemma) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Gemma3ForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Gemma3ForConditionalGeneration · 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 translategemma-12b-it: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 29 GB · Single high-end GPU (<80 GB) |
|---|---|
| FP8 (1 byte/param) | 15 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 7.3 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
finetune: 7quantized: 33adapter: 5merge: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| chbae624/vllm-translategemma-12b-it | finetune | — | 13.2B | L | gemma | chbae624 | 4 | 301.6k |
| mradermacher/translategemma-12b-it-GGUF | quantized | GGUF | — | — | gemma | mradermacher | 21 | 107.1k |
| kaitchup/translategemma-12b-it-NVFP4 | quantized | NVFP4 | 8.49B | M | gemma | kaitchup | 1 | 98.5k |
| bullerwins/translategemma-12b-it-GGUF | quantized | GGUF | — | — | gemma | bullerwins | 11 | 62.2k |
| mradermacher/translategemma-12b-it-i1-GGUF | quantized | GGUF | — | — | gemma | mradermacher | 10 | 23.9k |
| NikolayKozloff/translategemma-12b-it-Q4_K_M-GGUF | quantized | GGUF | — | — | gemma | NikolayKozloff | 2 | 16k |
| tatsuyaaaaaaa/translategemma-12b-it-gguf | quantized | gguf | — | — | — | tatsuyaaaaaaa | 0 | 9.4k |
| mlx-community/translategemma-12b-it-8bit | quantized | -8bit | 11.8B | L | gemma | mlx-community | 8 | 8.6k |
| mlx-community/translategemma-12b-it-4bit | quantized | -4bit | 11.8B | L | gemma | mlx-community | 6 | 8.1k |
| NikolayKozloff/translategemma-12b-it-Q5_K_M-GGUF | quantized | GGUF | — | — | gemma | NikolayKozloff | 2 | 6.5k |
| kaitchup/translategemma-12b-it-FP8-Dynamic | quantized | -FP8 | 12.2B | L | gemma | kaitchup | 3 | 3.3k |
| gateremark/kikuyu_translategemma_12b_merged_V2 | adapter | — | 13.2B | L | apache-2.0 | gateremark | 5 | 3k |
| steampunque/translategemma-12b-it-MP-GGUF | quantized | GGUF | — | — | gemma | steampunque | 0 | 3.3k |
| NikolayKozloff/translategemma-12b-it-Q8_0-GGUF | quantized | Q8_0 | — | — | gemma | NikolayKozloff | 4 | 2.8k |
| 42ailab/TranslateGemma-12B-GGUF | quantized | GGUF | — | — | gemma | 42ailab | 1 | 1.8k |
| NikolayKozloff/translategemma-12b-it-Q6_K-GGUF | quantized | Q6_K | — | — | gemma | NikolayKozloff | 3 | 1.5k |
| Sunbird/translategemma-12b-ug40 | finetune | — | 13.2B | L | apache-2.0 | Sunbird | 0 | 1k |
| Yacinedh/translategemma-12b-khutbah | quantized | GGUF | 12.2B | L | gemma | Yacinedh | 0 | 1k |
| DeltaKX/translategemma-12b-it-Q4_K_M-GGUF | quantized | GGUF | — | — | gemma | DeltaKX | 0 | 821 |
| anilb/translategemma-gguf | quantized | gguf | — | — | gemma | anilb | 0 | 755 |
| bRadu/translategemma-12b-it-novision | finetune | — | 11.8B | L | other | bRadu | 2 | 596 |
| mlx-community/translategemma-12b-it-6bit | quantized | — | 11.8B | L | gemma | mlx-community | 0 | 663 |
| NikolayKozloff/translategemma-12b-it-Q4_K_S-GGUF | quantized | GGUF | — | — | gemma | NikolayKozloff | 1 | 608 |
| piotreknow02/translategemma-12b-it-mlx-6bit | quantized | mlx | 11.8B | L | gemma | piotreknow02 | 0 | 549 |
| NikolayKozloff/translategemma-12b-it-Q5_K_S-GGUF | quantized | GGUF | — | — | gemma | NikolayKozloff | 1 | 489 |
| Abu-Dju/translategemma-12b-it-Q8_0-GGUF | quantized | Q8_0 | — | — | gemma | Abu-Dju | 0 | 493 |
| Rashik24/translategemma-12b-it-4bit | quantized | -4bit | 11.8B | L | gemma | Rashik24 | 0 | 475 |
| gateremark/kikuyu_translategemma_12b_merged | finetune | — | 13.2B | L | apache-2.0 | gateremark | 7 | 104 |
| async0x42/translategemma-12b-it-4bit | quantized | -4bit | 11.8B | L | gemma | async0x42 | 0 | 343 |
| SpongeBOB9684/GemmaTranslate-v3-12B | merge | — | 12.2B | L | gemma | SpongeBOB9684 | 2 | 226 |
| Riyan200324200324/translategemma-12b-it-GGUF | quantized | GGUF | — | — | gemma | Riyan200324200324 | 0 | 263 |
| litert-community/TranslateGemma-12B-IT | finetune | — | — | — | gemma | litert-community | 5 | 0 |
| Yacinedh/translategemma-12b-khutbah-lora | adapter | — | — | — | gemma | Yacinedh | 0 | 246 |
| wws11/parallel-chinese-translation-gguf-Q8_0 | quantized | gguf | — | — | — | wws11 | 0 | 229 |
| Abu-Dju/translategemma-12b-it-Q5_K_M-GGUF | quantized | GGUF | — | — | gemma | Abu-Dju | 0 | 209 |
| lab-ii/translategemma-12b-it-sah | finetune | — | 12.2B | L | gemma | lab-ii | 0 | 137 |
| kungaa/TuvanGemma | quantized | GGUF | — | — | other | kungaa | 0 | 136 |
| nufrnd/lvc-translategemma-12b-mlx | quantized | mlx | 11.8B | L | gemma | nufrnd | 0 | 91 |
| henry1477/translategemma-12b-it-NVFP4A16 | quantized | — | 11.8B | L | gemma | henry1477 | 0 | 77 |
| henry1477/translategemma-12b-it-NVFP4 | quantized | NVFP4 | 11.8B | L | gemma | henry1477 | 0 | 76 |
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.
| Title | Source | Published | First seen | Signals |
|---|---|---|---|---|
| Fine-Tuning TranslateGemma-4B for Better Welsh Translations on an H200 GPU | metalglot.com | 2026-05-13 | 2026-10-07 | hn |
| Comparing the best open source TranslateGemma projects | metalglot.com | 2026-05-03 | 2026-10-07 | hn |
| TranslateGemma: A new suite of open translation models | blog.google official | 2026-01-15 | 2026-10-07 | hn |
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 21:40 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:12 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.
