sam860/VibeThinker:1.5b-Q8_0

sam860/ VibeThinker:1.5b-Q8_0

221 Downloads Updated 2 months ago

tools

ollama run sam860/VibeThinker:1.5b-Q8_0

curl http://localhost:11434/api/chat \
  -d '{
    "model": "sam860/VibeThinker:1.5b-Q8_0",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='sam860/VibeThinker:1.5b-Q8_0',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'sam860/VibeThinker:1.5b-Q8_0',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 2 months ago

2 months ago

dc49be3809f3 · 1.9GB ·

model

archqwen2

·

parameters1.78B

·

quantizationQ8_0

1.9GB

template

{{- if .Suffix }}<|fim_prefix|>{{ .Prompt }}<|fim_suffix|>{{ .Suffix }}<|fim_middle|> {{- else if .M

1.6kB

params

{ "num_predict": 40960, "temperature": 0.6, "top_k": -1, "top_p": 0.95 }

64B

Readme

Notes

Uploaded in fp16 (full‑precision) and Q8_0 formats. Q8_0 is the default – it strikes the perfect balance for CPU inference while preserving nearly all of the fp16 quality.

Temperature: The model was tuned for deterministic math reasoning. Start with 0.6 (or 1.0 for more exploratory code generation). Lower values (≈0.2) can be used for very short fact‑lookup prompts.

Description

VibeThinker‑1.5B – a 1.5 B‑parameter dense model built on top of Qwen2.5‑Math‑1.5B.

Core innovation: Spectrum‑to‑Signal Principle (SSP) – a two‑stage training pipeline that first maximizes solution diversity during SFT, then reinforces correct signals with RL. This diversity‑first approach lets a tiny model punch far above its parameter count.
Specialty: Competitive‑style math (AIME, HMMT) and algorithmic coding (LeetCode, Codeforces). The model performs best when questions are asked in English.
Architecture: Standard decoder‑only transformer stack (dense) with a focus on efficient attention; no MoE or exotic layers.
Use‑case focus:
- Hard math problem solving
- Algorithmic code generation / reasoning
- Structured JSON output for tool‑calling (if needed)

Not recommended for general‑purpose chat, summarization, or creative writing.

References

VibeThinker GitHub

ModelScope page

Technical Report (arXiv 2511.06221)

Model Card on HuggingFace

### Notes
Uploaded in **fp16** (full‑precision) and **Q8_0** formats. **Q8_0 is the default** – it strikes the perfect balance for CPU inference while preserving nearly all of the fp16 quality.

**Temperature:** The model was tuned for deterministic math reasoning. Start with **0.6** (or **1.0** for more exploratory code generation). Lower values (≈0.2) can be used for very short fact‑lookup prompts.

---

### Description
**VibeThinker‑1.5B** – a 1.5 B‑parameter dense model built on top of **Qwen2.5‑Math‑1.5B**.

- **Core innovation:** *Spectrum‑to‑Signal Principle* (SSP) – a two‑stage training pipeline that first maximizes solution diversity during SFT, then reinforces correct signals with RL. This diversity‑first approach lets a tiny model punch far above its parameter count.  
- **Specialty:** Competitive‑style math (AIME, HMMT) and algorithmic coding (LeetCode, Codeforces). The model performs best when questions are asked in English.  
- **Architecture:** Standard decoder‑only transformer stack (dense) with a focus on efficient attention; no MoE or exotic layers.  
- **Use‑case focus:**  
  - Hard math problem solving  
  - Algorithmic code generation / reasoning  
  - Structured JSON output for tool‑calling (if needed)

*Not recommended* for general‑purpose chat, summarization, or creative writing.

---

### References
[VibeThinker GitHub](https://github.com/WeiboAI/VibeThinker)

[ModelScope page](https://modelscope.cn/models/WeiboAI/VibeThinker-1.5B)

[Technical Report (arXiv 2511.06221)](https://arxiv.org/abs/2511.06221)

[Model Card on HuggingFace](https://huggingface.co/WeiboAI/VibeThinker-1.5B)

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)