Text Generation
Transformers
Safetensors
Polish
gpt2
polish
nanogpt
tokenizer-ablation
research
Eval Results (legacy)
text-generation-inference
Instructions to use SlayerLab/GoLLeM-45M-PL with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use SlayerLab/GoLLeM-45M-PL with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="SlayerLab/GoLLeM-45M-PL")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("SlayerLab/GoLLeM-45M-PL") model = AutoModelForCausalLM.from_pretrained("SlayerLab/GoLLeM-45M-PL", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use SlayerLab/GoLLeM-45M-PL with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "SlayerLab/GoLLeM-45M-PL" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SlayerLab/GoLLeM-45M-PL", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/SlayerLab/GoLLeM-45M-PL
- SGLang
How to use SlayerLab/GoLLeM-45M-PL with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "SlayerLab/GoLLeM-45M-PL" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SlayerLab/GoLLeM-45M-PL", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "SlayerLab/GoLLeM-45M-PL" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SlayerLab/GoLLeM-45M-PL", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use SlayerLab/GoLLeM-45M-PL with Docker Model Runner:
docker model run hf.co/SlayerLab/GoLLeM-45M-PL
card: dolozona niezalezna weryfikacja SlayerLab (repro 1:1 bench_pl + kontekst 110M-v3), tresc autorki nietknieta
Browse files
README.md
CHANGED
|
@@ -190,3 +190,24 @@ Bazuje na [nanoGPT](https://github.com/karpathy/nanoGPT) (MIT, Andrej Karpathy).
|
|
| 190 |
url = {https://huggingface.co/KateMajzel/GoLLeM-45M-PL}
|
| 191 |
}
|
| 192 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 190 |
url = {https://huggingface.co/KateMajzel/GoLLeM-45M-PL}
|
| 191 |
}
|
| 192 |
```
|
| 193 |
+
|
| 194 |
+
---
|
| 195 |
+
|
| 196 |
+
## Weryfikacja niezależna (SlayerLab, 2026-08-30)
|
| 197 |
+
|
| 198 |
+
Zespół SlayerLab odtworzył wyniki z tej karty **niezależnie**, używając oryginalnego
|
| 199 |
+
harnessu autorki (`bench_pl.py` z [`KateMajzel/gollem-pl`](https://github.com/KateMajzel/gollem-pl),
|
| 200 |
+
domain-PMI, `--n 400 --seed 42`, bez modyfikacji):
|
| 201 |
+
|
| 202 |
+
| zadanie | karta (autorka) | nasza reprodukcja |
|
| 203 |
+
|---|---|---|
|
| 204 |
+
| PolEmo2-IN (acc PMI) | 47,2% | **47,2%** |
|
| 205 |
+
| 8Tags (acc PMI) | 31,5% | **31,5%** |
|
| 206 |
+
|
| 207 |
+
Zgodność 1:1 — wyniki w pełni reprodukowalne.
|
| 208 |
+
|
| 209 |
+
Dla kontekstu, na tym samym harnessie nasz model
|
| 210 |
+
[`SlayerLab/GoLLeM-110M-PL-v3`](https://huggingface.co/SlayerLab/GoLLeM-110M-PL-v3) (110M)
|
| 211 |
+
osiąga polemo2 56,5% / 8tags 37,2% (acc PMI) — zgodnie z oczekiwaniem dla większej skali.
|
| 212 |
+
Uwaga autorki pozostaje w mocy: modele tej wielkości bywają blisko poziomu losowego na
|
| 213 |
+
raw-acc, a sygnał wyciąga dopiero normalizacja PMI.
|