Instructions to use dblakely/WizardLM-13B-V1.2-fixed-tokenizer with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dblakely/WizardLM-13B-V1.2-fixed-tokenizer with Transformers:

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

pipe = pipeline("text-generation", model="dblakely/WizardLM-13B-V1.2-fixed-tokenizer")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("dblakely/WizardLM-13B-V1.2-fixed-tokenizer")
model = AutoModelForCausalLM.from_pretrained("dblakely/WizardLM-13B-V1.2-fixed-tokenizer")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use dblakely/WizardLM-13B-V1.2-fixed-tokenizer with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "dblakely/WizardLM-13B-V1.2-fixed-tokenizer"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "dblakely/WizardLM-13B-V1.2-fixed-tokenizer",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/dblakely/WizardLM-13B-V1.2-fixed-tokenizer

SGLang

How to use dblakely/WizardLM-13B-V1.2-fixed-tokenizer 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 "dblakely/WizardLM-13B-V1.2-fixed-tokenizer" \
    --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": "dblakely/WizardLM-13B-V1.2-fixed-tokenizer",
		"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 "dblakely/WizardLM-13B-V1.2-fixed-tokenizer" \
        --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": "dblakely/WizardLM-13B-V1.2-fixed-tokenizer",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use dblakely/WizardLM-13B-V1.2-fixed-tokenizer with Docker Model Runner:
```
docker model run hf.co/dblakely/WizardLM-13B-V1.2-fixed-tokenizer
```

dblakely commited on Jan 22, 2024

Commit

abaefe7

verified ·

1 Parent(s): 52ac67b

Update README.md

Browse files

Files changed (1) hide show

README.md +7 -0

README.md CHANGED Viewed

@@ -1,3 +1,10 @@
 ---
 license: llama2
 ---

 ---
 license: llama2
 ---
+ This is a slightly modified versions of the original [WizardLM/WizardLM-13B-V1.2 checkpoint](https://huggingface.co/WizardLM/WizardLM-13B-V1.2) that fixes a few bugs:
+* In the original checkpoint, the BOS token is set to the EOS token (`</s>`, token ID 2). In this version, the BOS is reverted to `<s>` (token ID 1).
+* The original has a mismatch between the size of the tokenizer vocab and the model embedding vocab. This is because the tokenizer includes an extra token for the added `[PAD]` token, making the vocab 32,001 tokens. This discrepancy can cause index errors. This version simply removes the added `[PAD]` in favor of using the `<unk>` (token ID 0) for padding. So the tokenizer's vocab is reverted back to a size of 32,000 to match the model's vocab size.
+For all other information about this model, refer to the original [WizardLM/WizardLM-13B-V1.2 checkpoint](https://huggingface.co/WizardLM/WizardLM-13B-V1.2).