Instructions to use TaylorAI/Flash-Llama-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use TaylorAI/Flash-Llama-3B with Transformers:

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

pipe = pipeline("text-generation", model="TaylorAI/Flash-Llama-3B", trust_remote_code=True)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("TaylorAI/Flash-Llama-3B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("TaylorAI/Flash-Llama-3B", trust_remote_code=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use TaylorAI/Flash-Llama-3B with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/TaylorAI/Flash-Llama-3B

SGLang

How to use TaylorAI/Flash-Llama-3B 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 "TaylorAI/Flash-Llama-3B" \
    --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": "TaylorAI/Flash-Llama-3B",
		"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 "TaylorAI/Flash-Llama-3B" \
        --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": "TaylorAI/Flash-Llama-3B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use TaylorAI/Flash-Llama-3B with Docker Model Runner:
```
docker model run hf.co/TaylorAI/Flash-Llama-3B
```

andersonbcdefg commited on Aug 25, 2023

Commit

87cede6

1 Parent(s): 22143dc

Upload modeling_flash_llama.py

Browse files

Files changed (1) hide show

modeling_flash_llama.py +10 -0

modeling_flash_llama.py CHANGED Viewed

@@ -369,6 +369,12 @@ class LlamaAttention(nn.Module):
             unpadded_kv, indices_k, cu_seqlens_k, max_seqlen_k = unpad_input(kv, attention_mask)
             unpadded_q, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(q, attention_mask[:, -q.size(1):])
             attn_outputs = flash_attn_varlen_kvpacked_func(
                 unpadded_q, unpadded_kv, cu_seqlens_q, cu_seqlens_k,
                 max_seqlen_q, max_seqlen_k,
@@ -385,6 +391,10 @@ class LlamaAttention(nn.Module):
         else:
             # no padding tokens, more efficient
             attn_outputs = flash_attn_kvpacked_func(
                 q, kv, dropout_p=0.0, softmax_scale=1.0/self.norm_factor, causal=(not has_layer_past), return_attn_probs=output_attentions)

             unpadded_kv, indices_k, cu_seqlens_k, max_seqlen_k = unpad_input(kv, attention_mask)
             unpadded_q, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(q, attention_mask[:, -q.size(1):])
+            # cast kv and q to bf16 or fp16 if currently in float32
+            if unpadded_kv.dtype == torch.float32:
+                unpadded_kv = unpadded_kv.to(torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16)
+                unpadded_q = unpadded_q.to(torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16)
             attn_outputs = flash_attn_varlen_kvpacked_func(
                 unpadded_q, unpadded_kv, cu_seqlens_q, cu_seqlens_k,
                 max_seqlen_q, max_seqlen_k,
         else:
             # no padding tokens, more efficient
+            # cast to bf16 or fp16 if currently in float32
+            if kv.dtype == torch.float32:
+                kv = kv.to(torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16)
+                q = q.to(torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16)
             attn_outputs = flash_attn_kvpacked_func(
                 q, kv, dropout_p=0.0, softmax_scale=1.0/self.norm_factor, causal=(not has_layer_past), return_attn_probs=output_attentions)