Downloads · 30 days
55
4% of all-time downloads
Gros-Michel-90m-Base-v2
finnianx/Gros-Michel-90m-Base-v2
Gros-Michel-90m-Base-v2 is a machine learning model from finnianx. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Gros-Michel-90m-Base-v2 is a 95m parameter billingual LLM trained on 9 billion tokens of a custom dataset mixture, with a new custom 25k tokenizer focusing on english and german and a 1024 token context length. The go…
All-time downloads
1.3K
Public
Parameters
94.9M
190 MB on disk
Likes
6
Public
Repo makeup
Click a slice to open those files.
.safetensors190 MB · 99%
At a glance
- Model type
- llama
- Access
- Public
- Created
- Jul 5, 2026
- Updated
- Jul 10, 2026
- SHA
- 1ccece3a
Datasets
Model card
From the Hugging Face model README
Gros-Michel-90m-Base-v2 is a 95m parameter billingual LLM trained on 9 billion tokens of a custom dataset mixture, with a new custom 25k tokenizer focusing on english and german and a 1024 token context length. The goal with this model is to provide a flexible base for further finetuning on downstream tasks, such as translation, sentiment analysis and extraction. This is the second edition in the Gros-Michel series, and shows significant improvements in almost all benchmarks.
Pretrain Data mixture
| Dataset | Weight |
|---|---|
HuggingFaceFW/fineweb-edu | 40% |
epfml/FineWeb-HQ | 16% |
HuggingFaceTB/cosmopedia (stories split) | 2% |
HuggingFaceTB/finemath (finemath-4plus) | 2% |
LSX-UniWue/LLaMmlein-Dataset | 40% |
Comparison to other models
| Maker | Model | Hellaswag | ARC (easy) | PIQA | BLiMP | Average |
|---|---|---|---|---|---|---|
| finnianx | Gros-Michel-90M-v2 | 31.26% | 43.18% | 60.77% | 80.20% | 53.85% |
| finnianx | Gros-Michel-90M | 30.26% | 41.50% | 59.41% | 78.35% | 52.38% |
| MaliosDark | Isabel-50M | 27.1% | 43.81% | 57.12% | 73.75% | 50.44% |
| finnianx | Michel-Nano-v2 | 27.40% | 35.90% | 56.75% | 72.52% | 48.14% |
| Axiomic Labs | GPT-S-5M | 27.39% | 33.16% | 57.13% | 72.21% | 47.47% |
| EleutherAI | pythia-31m | 27.14% | 33.88% | 56.26% | 67.78% | 46.27% |
German Benchmarks
| Model | arc_de acc | arc_de acc_norm | hellaswag_de acc | hellaswag_de acc_norm | m_mmlu_de acc | truthfulqa_de_mc1 acc | truthfulqa_de_mc2 acc |
|---|---|---|---|---|---|---|---|
| Gros-Michel-90M-Base-v2 | 0.1925 | 0.2310 | 0.2816 | 0.2929 | 0.2332 | 0.2449 | 0.4287 |
| Gros-Michel-90M-Base | 0.1865 | 0.2284 | 0.2697 | 0.2852 | 0.2346 | 0.2348 | 0.4285 |
| nanochat German v1 | 0.2241 | 0.2626 | 0.3203 | 0.3581 | 0.2285 | 0.2500 | 0.4184 |
| LLäMmlein-120M | 0.1942 | 0.2301 | 0.2945 | 0.3178 | 0.2285 | 0.2310 | 0.4055 |
| LLäMmlein-1B | 0.2515 | 0.2960 | 0.3703 | 0.4490 | 0.2317 | 0.2322 | 0.3617 |
Notice
This model has not undergone any alignment, and therefore may produce harmful content.
Evaluation was done in lm-eval-harness by EleutherAI, all benchmark scores use normalized accuracy where applicable and are zero-shot. Weights are stored in BF16.
Future plans
Sometime in the near(ish) future i will release an instruction tuned variant of this model, along with a translation focused finetune. GGUF support will also come in the near(ish) future.
- Type
- llama
- Languages
- de, en
- Created
- Jul 5, 2026
- Updated
- Jul 10, 2026