113 lines
2.5 KiB
Markdown
113 lines
2.5 KiB
Markdown
---
|
|
license: apache-2.0
|
|
datasets:
|
|
- NousResearch/Hermes-3-Dataset
|
|
language:
|
|
- en
|
|
base_model:
|
|
- Qwen/Qwen3-4B-Thinking-2507
|
|
---
|
|
|
|
# Qwen3-4B-Thinking-2507-Hermes-3
|
|
A qwen3 4b 2507 thinking model finetuned with the hermes 3 dataset.
|
|
|
|
## Capabilities:
|
|
- Reasoning retained
|
|
- Better instruction following
|
|
|
|
### How to run:
|
|
|
|
#### Transformers
|
|
Run this code
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3"
|
|
|
|
# load the tokenizer and the model
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto"
|
|
)
|
|
|
|
# prepare the model input
|
|
prompt = "Give me a short introduction to large language model."
|
|
messages = [
|
|
{"role": "user", "content": prompt}
|
|
]
|
|
text = tokenizer.apply_chat_template(
|
|
messages,
|
|
tokenize=False,
|
|
add_generation_prompt=True,
|
|
)
|
|
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
|
|
|
# conduct text completion
|
|
generated_ids = model.generate(
|
|
**model_inputs,
|
|
max_new_tokens=32768
|
|
)
|
|
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
|
|
|
|
# parsing thinking content
|
|
try:
|
|
# rindex finding 151668 (</think>)
|
|
index = len(output_ids) - output_ids[::-1].index(151668)
|
|
except ValueError:
|
|
index = 0
|
|
|
|
thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
|
|
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")
|
|
|
|
print("thinking content:", thinking_content) # no opening <think> tag
|
|
print("content:", content)
|
|
```
|
|
|
|
#### Vllm
|
|
Run this command
|
|
```bash
|
|
vllm serve ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3 --max-model-len 262144 --enable-reasoning --reasoning-parser deepseek_r1
|
|
```
|
|
|
|
#### Sglang
|
|
Run this command
|
|
```bash
|
|
python -m sglang.launch_server --model-path ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3 --context-length 262144 --reasoning-parser deepseek-r1
|
|
```
|
|
|
|
#### Llama.cpp
|
|
Run this command
|
|
```
|
|
llama-server --hf-repo ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3
|
|
```
|
|
|
|
#### Ollama
|
|
Run this command
|
|
```bash
|
|
ollama run hf.co/ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3:IQ4_NL
|
|
```
|
|
or
|
|
```bash
|
|
ollama run hf.co/ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3:Q5_K_M
|
|
```
|
|
|
|
#### Lm Studio
|
|
[LM Studio link](https://model.lmstudio.ai/download/ertghiu256/Qwen3-Hermes-4b)
|
|
|
|
### Recommended parameters
|
|
```
|
|
Temp: 0.6
|
|
Top_P: 20
|
|
Top_K: 0.95
|
|
```
|
|
|
|
### Training details
|
|
```
|
|
Trained with Unsloth
|
|
Training parameters
|
|
- 60 steps
|
|
- 3-e5 Learning rate
|
|
- 28k samples from Hermes 3 dataset
|
|
``` |