Instructions to use poolside-laguna-hackathon/laguna-xs2-dense-stage2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use poolside-laguna-hackathon/laguna-xs2-dense-stage2 with Transformers:

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

pipe = pipeline("text-generation", model="poolside-laguna-hackathon/laguna-xs2-dense-stage2", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("poolside-laguna-hackathon/laguna-xs2-dense-stage2", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("poolside-laguna-hackathon/laguna-xs2-dense-stage2", trust_remote_code=True, device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use poolside-laguna-hackathon/laguna-xs2-dense-stage2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "poolside-laguna-hackathon/laguna-xs2-dense-stage2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "poolside-laguna-hackathon/laguna-xs2-dense-stage2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/poolside-laguna-hackathon/laguna-xs2-dense-stage2

SGLang

How to use poolside-laguna-hackathon/laguna-xs2-dense-stage2 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 "poolside-laguna-hackathon/laguna-xs2-dense-stage2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "poolside-laguna-hackathon/laguna-xs2-dense-stage2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "poolside-laguna-hackathon/laguna-xs2-dense-stage2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "poolside-laguna-hackathon/laguna-xs2-dense-stage2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use poolside-laguna-hackathon/laguna-xs2-dense-stage2 with Docker Model Runner:
```
docker model run hf.co/poolside-laguna-hackathon/laguna-xs2-dense-stage2
```

duico commited on May 30

Commit

daf8bea

verified ·

1 Parent(s): 852aaa4

Upload README.md with huggingface_hub

Browse files

Files changed (1) hide show

README.md +11 -12

README.md CHANGED Viewed

@@ -55,24 +55,23 @@ print(tok.decode(m.generate(**ids, max_new_tokens=64)[0], skip_special_tokens=Tr
 ## Results
-| Model | Params (resident) | Held-out PPL | HumanEval pass@1 |
-|---|---|---|---|
-| Teacher (Laguna XS.2, fp8) | 33B (3B active) | ≈4.4 | _running_ |
-| [Stage-1 dense](https://huggingface.co/poolside-laguna-hackathon/laguna-xs2-dense-stage1) | ≈3B | ≈25 | 0.0% (0/164) |
-| **Stage-2 dense (this)** | ≈3B | _tbd_ | **0.6%** (1/164) |
-_(HumanEval pass@1 via [evalplus](https://github.com/evalplus/evalplus), greedy. Teacher number being measured; will be filled in.)_
-Both dense checkpoints are **near-zero on HumanEval** — expected at this tiny distillation budget; the model is **not yet usable** for code generation. What this run demonstrates is the **architecture + footprint** (below) and a **directional** improvement (Stage-1 0.0% → Stage-2 0.6%, KL 2.5 → 1.40). Closing the task-accuracy gap needs much more KD.
-### Sample generations
-On a simple completion prompt — `def fibonacci(n): """Return the n-th Fibonacci number."""` (greedy) — both models are still degenerate, but Stage-2 is visibly less broken:
-- **Stage-1:** non-Pythonic repetition — `return n;` repeated indefinitely.
-- **Stage-2:** *valid* Python with the correct function structure, but stuck in a loop — `return 0` repeated.
-The KD moves the model directionally toward coherent code; it just needs far more than 14M tokens.
 ## VRAM / footprint

 ## Results
+HumanEval pass@1 (greedy, [evalplus](https://github.com/evalplus/evalplus)), base / plus:
+| Model | Params (resident) | PPL | HumanEval (raw completion) | HumanEval (chat template) |
+|---|---|---|---|---|
+| Teacher (Laguna XS.2, fp8) | 33B (3B active) | ≈4.4 | — | **88.4% / 84.8%** |
+| [Stage-1 dense](https://huggingface.co/poolside-laguna-hackathon/laguna-xs2-dense-stage1) | ≈3B | ≈25 | 0.0% | 0.0% |
+| **Stage-2 dense (this)** | ≈3B | — | 0.6% | 0.0% |
+## Diagnosis & next steps (honest)
+The dense models are **near-zero on HumanEval in *both* eval formats** — i.e. genuinely **not yet usable**, not just an eval artifact. The teacher scores a normal **88.4%** with its chat template, so the harness is sound and the eval format matters a lot (raw completion is off-distribution for this instruct model).
+**Root cause:** XS.2 is an **instruct/agentic** model (chat template, special tokens, EOS-terminated turns), but we distilled it on **raw concatenated pretraining text** (DCLM + code, packed). That pushed the student off its instruct distribution and it never learned to *stop* — so generations degenerate (Stage-1: `return n;` repeated; Stage-2 in chat mode: control-token spam `</think>…</assistant>`). The ≈14M-token KD budget is also 20–400× below typical recovery budgets.
+**Fix (the real next step):** distill in the model's **native chat format** — coding instruction→response conversations rendered through `chat_template.jinja`, EOS-terminated, ideally with teacher-generated responses (so we match the teacher's actual operating distribution). Same KD loss/loop; only the data + tokenization change. More *raw* tokens would not fix this.
+What this release **does** demonstrate: the MoE→dense **architecture** works (per-layer init converges, see the NMSE curves) and the **≈11× weight-VRAM reduction** (below) at matched active-compute. Task accuracy awaits the chat-format KD run.
 ## VRAM / footprint