Instructions to use RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds with Transformers:

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

pipe = pipeline("text-generation", model="RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds")
model = AutoModelForCausalLM.from_pretrained("RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds

SGLang

How to use RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds 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 "RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds" \
    --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": "RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds",
		"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 "RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds" \
        --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": "RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds with Docker Model Runner:
```
docker model run hf.co/RedHatAI/MiniChat-1.5-3B-pruned50-quant-ds
```

mwitiderrick commited on Dec 6, 2023

Commit

df82913

1 Parent(s): b28eb0a

Create recipe.yaml

Browse files

Files changed (1) hide show

recipe.yaml +37 -0

recipe.yaml ADDED Viewed

	@@ -0,0 +1,37 @@

+test_stage:
+  obcq_modifiers:
+    SmoothQuantModifier:
+      smoothing_strength: 0.8
+      mappings: [
+        [["re:.*q_proj", "re:.*k_proj", "re:.*v_proj"], "re:.*input_layernorm"],
+        [["re:.*gate_proj", "re:.*up_proj"], "re:.*post_attention_layernorm"]
+      ]
+    QuantizationModifier:
+      ignore:
+      # These operations don't make sense to quantize
+      - LlamaRotaryEmbedding
+      - LlamaRMSNorm
+      - SiLUActivation
+      # Skip quantizing the BMMs
+      - QuantizableMatMul
+      # Skip quantizing the layers with the most sensitive activations
+      - model.layers.21.mlp.down_proj
+      - model.layers.7.mlp.down_proj
+      - model.layers.2.mlp.down_proj
+      - model.layers.20.mlp.down_proj
+      - model.layers.19.mlp.down_proj
+      post_oneshot_calibration: true
+      scheme_overrides:
+        Embedding:
+          input_activations: null
+          weights:
+            num_bits: 8
+            symmetric: false
+    SparseGPTModifier:
+      sparsity: 0.5
+      block_size: 128
+      sequential_update: true
+      quantize: true
+      percdamp: 0.01
+      mask_structure: "0:0"
+      targets: ["re:model.layers.\\d*$"]