Instructions to use tiiuae/falcon-40b-instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiiuae/falcon-40b-instruct with Transformers:

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

pipe = pipeline("text-generation", model="tiiuae/falcon-40b-instruct", trust_remote_code=True)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-40b-instruct", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("tiiuae/falcon-40b-instruct", trust_remote_code=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tiiuae/falcon-40b-instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tiiuae/falcon-40b-instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tiiuae/falcon-40b-instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/tiiuae/falcon-40b-instruct

SGLang

How to use tiiuae/falcon-40b-instruct 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 "tiiuae/falcon-40b-instruct" \
    --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": "tiiuae/falcon-40b-instruct",
		"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 "tiiuae/falcon-40b-instruct" \
        --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": "tiiuae/falcon-40b-instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use tiiuae/falcon-40b-instruct with Docker Model Runner:
```
docker model run hf.co/tiiuae/falcon-40b-instruct
```

Use input attention mask instead of casual mask in attention

#72

by CyberZHG - opened Aug 15, 2023

base: refs/heads/main

←

from: refs/pr/72

Discussion Files changed

-2

Files changed (1) hide show

modelling_RW.py +2 -2

modelling_RW.py CHANGED Viewed

@@ -281,13 +281,14 @@ class Attention(nn.Module):
         else:
             present = None
         if alibi is None:
             query_layer_ = query_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
             key_layer_ = key_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
             value_layer_ = value_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
             attn_output = F.scaled_dot_product_attention(
-                query_layer_, key_layer_, value_layer_, None, 0.0, is_causal=True
             )
             x = attn_output.view(batch_size, self.num_heads, q_length, self.head_dim)
@@ -300,7 +301,6 @@ class Attention(nn.Module):
             assert not output_attentions  # not supported.
             return outputs
         else:
-            attention_mask_float = (attention_mask * 1.0).masked_fill(attention_mask, -1e9).to(torch.bfloat16)
             matmul_result = query_layer @ key_layer.transpose(-1, -2)
             # change view to [batch_size, num_heads, q_length, kv_length]

         else:
             present = None
+        attention_mask_float = (attention_mask * 1.0).masked_fill(attention_mask, -1e9).to(query_layer.dtype)
         if alibi is None:
             query_layer_ = query_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
             key_layer_ = key_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
             value_layer_ = value_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
             attn_output = F.scaled_dot_product_attention(
+                query_layer_, key_layer_, value_layer_, attention_mask_float, 0.0, is_causal=False
             )
             x = attn_output.view(batch_size, self.num_heads, q_length, self.head_dim)
             assert not output_attentions  # not supported.
             return outputs
         else:
             matmul_result = query_layer @ key_layer.transpose(-1, -2)
             # change view to [batch_size, num_heads, q_length, kv_length]