Instructions to use keepsloading/icml_repro_scratch with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use keepsloading/icml_repro_scratch with Transformers:

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

pipe = pipeline("text-generation", model="keepsloading/icml_repro_scratch")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("keepsloading/icml_repro_scratch")
model = AutoModelForCausalLM.from_pretrained("keepsloading/icml_repro_scratch", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use keepsloading/icml_repro_scratch with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/keepsloading/icml_repro_scratch

SGLang

How to use keepsloading/icml_repro_scratch 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 "keepsloading/icml_repro_scratch" \
    --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": "keepsloading/icml_repro_scratch",
		"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 "keepsloading/icml_repro_scratch" \
        --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": "keepsloading/icml_repro_scratch",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use keepsloading/icml_repro_scratch with Docker Model Runner:
```
docker model run hf.co/keepsloading/icml_repro_scratch
```

icml_repro_scratch / outputs /eval /token_t045 /sparsity_ruler_b.json

keepsloading

Upload folder using huggingface_hub

1ade3b4 verified 26 days ago

Raw

History Blame Contribute Delete

4.17 kB

	{
	"definition": "hard AHA gate zeros / all token x KV-head x layer decisions, token-weighted over prefill and decode",
	"router_granularity": "token",
	"native_router_decisions": 25272464,
	"effective_router_decisions": 202179712,
	"native_router_note": "Native decisions count learned gate logits before broadcast; effective decisions always use token x KV-head x layer for matched sparsity.",
	"sparse_decisions": 15539856,
	"total_decisions": 202179712,
	"sparsity": 0.07686159924889002,
	"full_attention_usage": 0.92313840075111,
	"per_layer": {
	"0": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"1": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"2": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"3": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"4": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"5": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"6": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"7": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"8": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"9": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"10": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"11": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"12": {
	"sparse_decisions": 0,
	"total_decisions": 7220704,
	"sparsity": 0.0
	},
	"13": {
	"sparse_decisions": 432,
	"total_decisions": 7220704,
	"sparsity": 5.982796137329546e-05
	},
	"14": {
	"sparse_decisions": 3544,
	"total_decisions": 7220704,
	"sparsity": 0.000490810868303146
	},
	"15": {
	"sparse_decisions": 378032,
	"total_decisions": 7220704,
	"sparsity": 0.05235389790247599
	},
	"16": {
	"sparse_decisions": 1952,
	"total_decisions": 7220704,
	"sparsity": 0.00027033375139044616
	},
	"17": {
	"sparse_decisions": 131768,
	"total_decisions": 7220704,
	"sparsity": 0.01824863614406573
	},
	"18": {
	"sparse_decisions": 33576,
	"total_decisions": 7220704,
	"sparsity": 0.004649962108957797
	},
	"19": {
	"sparse_decisions": 64112,
	"total_decisions": 7220704,
	"sparsity": 0.008878912637881292
	},
	"20": {
	"sparse_decisions": 123288,
	"total_decisions": 7220704,
	"sparsity": 0.017074235420812153
	},
	"21": {
	"sparse_decisions": 287688,
	"total_decisions": 7220704,
	"sparsity": 0.039842098498982924
	},
	"22": {
	"sparse_decisions": 1334864,
	"total_decisions": 7220704,
	"sparsity": 0.18486618479306172
	},
	"23": {
	"sparse_decisions": 3498784,
	"total_decisions": 7220704,
	"sparsity": 0.48454887501274113
	},
	"24": {
	"sparse_decisions": 1545728,
	"total_decisions": 7220704,
	"sparsity": 0.2140688774944936
	},
	"25": {
	"sparse_decisions": 860456,
	"total_decisions": 7220704,
	"sparsity": 0.11916511187828777
	},
	"26": {
	"sparse_decisions": 56296,
	"total_decisions": 7220704,
	"sparsity": 0.007796469707108892
	},
	"27": {
	"sparse_decisions": 7219336,
	"total_decisions": 7220704,
	"sparsity": 0.9998105447889846
	}
	},
	"by_phase": {
	"decode": {
	"sparse_decisions": 270424,
	"total_decisions": 2947840,
	"sparsity": 0.09173632218844985,
	"full_attention_usage": 0.9082636778115502
	},
	"prefill": {
	"sparse_decisions": 15269432,
	"total_decisions": 199231872,
	"sparsity": 0.07664151245840826,
	"full_attention_usage": 0.9233584875415918
	}
	}
	}