Instructions to use MoritzMo123/kfac-svd_lems_llama-3-8b_0.9 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MoritzMo123/kfac-svd_lems_llama-3-8b_0.9 with Transformers:

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

pipe = pipeline("text-generation", model="MoritzMo123/kfac-svd_lems_llama-3-8b_0.9", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("MoritzMo123/kfac-svd_lems_llama-3-8b_0.9", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use MoritzMo123/kfac-svd_lems_llama-3-8b_0.9 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MoritzMo123/kfac-svd_lems_llama-3-8b_0.9"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MoritzMo123/kfac-svd_lems_llama-3-8b_0.9",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/MoritzMo123/kfac-svd_lems_llama-3-8b_0.9

SGLang

How to use MoritzMo123/kfac-svd_lems_llama-3-8b_0.9 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 "MoritzMo123/kfac-svd_lems_llama-3-8b_0.9" \
    --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": "MoritzMo123/kfac-svd_lems_llama-3-8b_0.9",
		"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 "MoritzMo123/kfac-svd_lems_llama-3-8b_0.9" \
        --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": "MoritzMo123/kfac-svd_lems_llama-3-8b_0.9",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use MoritzMo123/kfac-svd_lems_llama-3-8b_0.9 with Docker Model Runner:
```
docker model run hf.co/MoritzMo123/kfac-svd_lems_llama-3-8b_0.9
```

unsloth/llama-3-8b — lems search — 0.9 target ratio

This model was compressed using kfac_svd with lems rank search starting from unsloth/llama-3-8b as base model. You may check out our publication and project page for details on kfac-svd and our LEMS rank search.

Compression Details

Metric	Value
Base Model	`unsloth/llama-3-8b`
Method	`kfac_svd`
Search Method	`lems`
Target Ratio	0.9
Compression Metric	`params`
Recommended Dtype	`float16`
Compressed Layers	77
Total Parameters	7,332,188,237

Usage

The checkpoint records its recommended dtype in config.json; no explicit torch_dtype argument should be needed with this remote-code wrapper. For standard Transformers models, torch_dtype="auto" is the portable fallback.

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "MoritzMo123/kfac-svd_lems_llama-3-8b_0.9",
    trust_remote_code=True,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("MoritzMo123/kfac-svd_lems_llama-3-8b_0.9")

inputs = tokenizer('Hello, ', return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Evaluation Results

Dataset	Perplexity
wikitext2	6.58
ptb	11.29
c4	11.11

Rank Allocation

Per-layer ranks (click to expand)

Layer	Rank
`model.layers.0.self_attn.k_proj`	360
`model.layers.0.self_attn.q_proj`	632
`model.layers.1.self_attn.k_proj`	760
`model.layers.1.self_attn.q_proj`	624
`model.layers.10.mlp.down_proj`	2184
`model.layers.10.mlp.gate_proj`	1824
`model.layers.10.self_attn.k_proj`	712
`model.layers.10.self_attn.q_proj`	768
`model.layers.11.mlp.down_proj`	2112
`model.layers.11.mlp.gate_proj`	1696
`model.layers.11.self_attn.k_proj`	768
`model.layers.11.self_attn.o_proj`	1464
`model.layers.11.self_attn.q_proj`	816
`model.layers.12.mlp.gate_proj`	1752
`model.layers.12.self_attn.q_proj`	672
`model.layers.13.mlp.gate_proj`	1856
`model.layers.13.self_attn.q_proj`	880
`model.layers.14.self_attn.q_proj`	864
`model.layers.15.self_attn.q_proj`	776
`model.layers.16.self_attn.k_proj`	712
`model.layers.16.self_attn.q_proj`	648
`model.layers.17.self_attn.k_proj`	776
`model.layers.17.self_attn.q_proj`	744
`model.layers.18.self_attn.k_proj`	752
`model.layers.18.self_attn.q_proj`	784
`model.layers.19.self_attn.k_proj`	704
`model.layers.19.self_attn.o_proj`	1072
`model.layers.19.self_attn.q_proj`	632
`model.layers.2.self_attn.o_proj`	1208
`model.layers.2.self_attn.q_proj`	912
`model.layers.20.self_attn.k_proj`	512
`model.layers.20.self_attn.o_proj`	856
`model.layers.20.self_attn.q_proj`	632
`model.layers.21.self_attn.o_proj`	1048
`model.layers.21.self_attn.q_proj`	672
`model.layers.22.self_attn.k_proj`	592
`model.layers.22.self_attn.o_proj`	1080
`model.layers.22.self_attn.q_proj`	704
`model.layers.23.self_attn.k_proj`	568
`model.layers.23.self_attn.o_proj`	952
`model.layers.23.self_attn.q_proj`	680
`model.layers.24.self_attn.k_proj`	520
`model.layers.24.self_attn.o_proj`	808
`model.layers.24.self_attn.q_proj`	616
`model.layers.25.self_attn.k_proj`	608
`model.layers.25.self_attn.o_proj`	984
`model.layers.25.self_attn.q_proj`	648
`model.layers.26.self_attn.k_proj`	400
`model.layers.26.self_attn.o_proj`	944
`model.layers.26.self_attn.q_proj`	640
`model.layers.27.self_attn.k_proj`	640
`model.layers.27.self_attn.o_proj`	904
`model.layers.27.self_attn.q_proj`	640
`model.layers.28.self_attn.k_proj`	744
`model.layers.28.self_attn.o_proj`	1112
`model.layers.28.self_attn.q_proj`	744
`model.layers.29.self_attn.o_proj`	648
`model.layers.29.self_attn.q_proj`	696
`model.layers.3.self_attn.q_proj`	872
`model.layers.30.self_attn.k_proj`	544
`model.layers.30.self_attn.o_proj`	896
`model.layers.30.self_attn.q_proj`	624
`model.layers.31.self_attn.k_proj`	664
`model.layers.31.self_attn.o_proj`	1168
`model.layers.31.self_attn.q_proj`	704
`model.layers.4.self_attn.o_proj`	1240
`model.layers.4.self_attn.q_proj`	880
`model.layers.5.self_attn.o_proj`	1144
`model.layers.5.self_attn.q_proj`	752
`model.layers.6.self_attn.q_proj`	688
`model.layers.7.self_attn.k_proj`	776
`model.layers.7.self_attn.q_proj`	744
`model.layers.8.mlp.gate_proj`	1840
`model.layers.8.self_attn.q_proj`	880
`model.layers.9.mlp.gate_proj`	1936
`model.layers.9.self_attn.o_proj`	1568
`model.layers.9.self_attn.q_proj`	824

Hydra Configuration Summary

Config Field	Value
Model	`unsloth/llama-3-8b`
SVD Method	`kfac_svd`
Search Method	`lems`
Compression Target	`0.9`
Target Metric	`params`
Calibration Dataset	`wikitext2`
Sequence Length	`2048`
Seed	`42`