---
language:
- en
library_name: transformers
pipeline_tag: text-generation
tags:
- cobalt
- cobalt-2
- valiant
- valiant-labs
- qwen
- qwen-3
- qwen-3-14b
- 14b
- math
- math-reasoning
- math-instruct
- reasoning
- problem-solving
- creative
- analytical
- expert
- rationality
- conversational
- chat
- instruct
base_model: Qwen/Qwen3-14B
datasets:
- zwhe99/DeepMath-103K
- sequelbox/Raiden-DeepSeek-R1
license: apache-2.0
---


**[Support our open-source dataset and model releases!](https://huggingface.co/spaces/sequelbox/SupportOpenSource)**


Cobalt 2 is a math and general reasoning specialist built on Qwen 3.
- Finetuned on high-difficulty problems from [the math-reasoning DeepMath dataset](https://huggingface.co/datasets/zwhe99/DeepMath-103K) generated with Deepseek R1!
- Improved [general and creative reasoning](https://huggingface.co/datasets/sequelbox/Raiden-DeepSeek-R1) to supplement problem-solving and general chat performance.
- Small model sizes allow running on local desktop and mobile, plus super-fast server inference!


Try Esper 3, our full-stack code, architecture, and DevOps assistant: [Qwen3-4B](https://huggingface.co/ValiantLabs/Qwen3-4B-Esper3), [Qwen3-8B](https://huggingface.co/ValiantLabs/Qwen3-8B-Esper3), [Qwen3-14B](https://huggingface.co/ValiantLabs/Qwen3-14B-Esper3)


## Prompting Guide
Cobalt 2 uses the [Qwen 3](https://huggingface.co/Qwen/Qwen3-14B) prompt format.

Cobalt 2 is a reasoning finetune; **we recommend enable_thinking=True for all chats.**

Example inference script to get started:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "ValiantLabs/Qwen3-14B-Cobalt2"

# 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 = "Evaluate the limit using the Central Limit Theorem: \[ \lim_{n\to\infty}p^{n}\sum_{k \geqslant{n(p^{-1}-1)}}^{\infty}\binom{n+k-1}{n-1}(1-p)^{k}. \]"
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True # Switches between thinking and non-thinking modes. Default is 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)
print("content:", content)
```


![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/63444f2687964b331809eb55/VCJ8Fmefd8cdVhXSSxJiD.jpeg)


Cobalt 2 is created by [Valiant Labs.](http://valiantlabs.ca/)

[Check out our HuggingFace page to see Esper 3 and all of our models!](https://huggingface.co/ValiantLabs)

We care about open source. For everyone to use.