Instructions to use SlayerLab/pollock-mini-lm-125m with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SlayerLab/pollock-mini-lm-125m with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="SlayerLab/pollock-mini-lm-125m", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("SlayerLab/pollock-mini-lm-125m", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use SlayerLab/pollock-mini-lm-125m with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SlayerLab/pollock-mini-lm-125m"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SlayerLab/pollock-mini-lm-125m",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/SlayerLab/pollock-mini-lm-125m

SGLang

How to use SlayerLab/pollock-mini-lm-125m 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/pollock-mini-lm-125m" \
    --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/pollock-mini-lm-125m",
		"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/pollock-mini-lm-125m" \
        --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/pollock-mini-lm-125m",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use SlayerLab/pollock-mini-lm-125m with Docker Model Runner:
```
docker model run hf.co/SlayerLab/pollock-mini-lm-125m
```

eval: niezalezna ewaluacja r003(v1.1) vs r006 (Arkadiusz Slota) - pliki, bez zmian karty

by Maggio33 - opened Aug 30

base: refs/heads/main

←

from: refs/pr/2

Discussion Files changed

+9380

-0

Files changed (3) hide show

evaluations/independent-eval-slota-2026-08-30.md +40 -0
evaluations/r003-v1.1.json +0 -0
evaluations/r006.json +0 -0

evaluations/independent-eval-slota-2026-08-30.md ADDED Viewed

	@@ -0,0 +1,40 @@

+# Niezależna ewaluacja — r003 (v1.1) vs r006 (v1.4)
+**Autor:** Arkadiusz Słota · **Data:** 2026-08-30 · **Kontrybucja:** dodana jako PR (karta modelu nietknięta)
+Rzetelne, powtarzalne testy zero-shot obu rewizji na **jednym harnessie**
+(`lm-evaluation-harness 0.4.12` — ta sama wersja co w karcie), bf16, batch 8, pełne splity,
+GPU AMD RX 7900 XTX (ROCm/WSL). Surowe wyniki: `evaluations/r003-v1.1.json`, `evaluations/r006.json`.
+## Walidacja setupu (nasz r006 vs liczby z karty)
+LAMBADA 0,2750/ppl 49,999 (karta 0,2769/49,907) · HellaSwag 0,3014 (0,3013) · SciQ 0,6630 (0,6640)
+· ARC-C 0,2440 (0,2415) · ARC-E 0,4314 (0,4343) · PIQA 0,5914 (0,5968) · BLiMP 0,7801 (0,7692).
+→ 5/7 tasków 1:1; BLiMP +0,011 (różnica wersji `datasets`). Setup potwierdzony.
+## Wynik: r003 (v1.1) vs r006 (v1.4) — identyczny harness
+| task | metryka | r003 (v1.1) | r006 (v1.4) | Δ (v1.1−r006) | stderr |
+|---|---|---:|---:|---:|---:|
+| BLiMP | acc | **0,7921** | 0,7801 | +0,0120 | 0,0014 |
+| LAMBADA | acc | **0,2926** | 0,2750 | +0,0176 | 0,0063 |
+| LAMBADA | ppl ↓ | **47,61** | 50,00 | −2,39 | ~1,8 |
+| PIQA | acc_norm | **0,6143** | 0,5914 | +0,0229 | 0,0114 |
+| HellaSwag | acc_norm | 0,3036 | 0,3014 | +0,0022 | 0,0046 |
+| ARC-Challenge | acc_norm | 0,2517 | 0,2440 | +0,0077 | 0,0126 |
+| SciQ | acc_norm | 0,6570 | 0,6630 | −0,0060 | 0,0150 |
+| ARC-Easy | acc_norm | 0,4272 | 0,4314 | −0,0042 | 0,0102 |
+## Wniosek
+**r003 (v1.1) ≥ r006 (v1.4) na 6 z 8 metryk; istotnie lepszy na BLiMP, LAMBADA (acc+ppl) i PIQA.**
+Rewizje r004–r006 (m.in. swap tokenizera w r005 i powrót w r006) nie podniosły downstream zero-shot —
+r003 pozostaje mocnym, miejscami najlepszym checkpointem.
+Zastrzeżenia: to zero-shot na modelu ~127M (liczy się różnica, nie poziom bezwzględny); zmiana
+v1.1→v1.4 miesza korpus + tokenizer + kroki, więc nie izoluje pojedynczej przyczyny — pełne „dlaczego"
+wymagałoby ablacji per-zmiana (ten sam korpus/tokenizer/budżet, jedna zmienna naraz).
+## Reprodukcja
+```
+lm_eval --model hf \
+  --model_args pretrained=SlayerLab/pollock-mini-lm-125m,revision=<v1.1|a2e53d9>,dtype=bfloat16 \
+  --tasks blimp,lambada_openai,hellaswag,piqa,sciq,arc_easy,arc_challenge --batch_size 8 --device cuda
+```

evaluations/r003-v1.1.json ADDED Viewed

The diff for this file is too large to render. See raw diff

evaluations/r006.json ADDED Viewed

The diff for this file is too large to render. See raw diff