初始化项目，由ModelHub XC社区提供模型
Model: OpceanAI/Yuuki-RxG-nano Source: Original Platform
This commit is contained in:
ModelHub XC
40
.gitattributes
vendored
Normal file
40
.gitattributes
vendored
Normal file
@@ -0,0 +1,40 @@
|
|||||||
|
*.7z filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.arrow filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.bin filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.ftz filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.gz filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.h5 filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.joblib filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.model filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.npy filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.npz filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.onnx filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.ot filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.parquet filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.pb filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.pickle filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.pkl filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.pt filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.pth filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.rar filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
||||||
|
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.tar filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.tflite filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.tgz filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.wasm filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.xz filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.zip filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*.zst filter=lfs diff=lfs merge=lfs -text
|
||||||
|
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
||||||
|
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
||||||
|
yuuki_rxg_nano_benchmarks_v5.png filter=lfs diff=lfs merge=lfs -text
|
||||||
|
yuuki_rxg_nano_benchmarks_png filter=lfs diff=lfs merge=lfs -text
|
||||||
|
yuuki_rxg_nano_benchmarkspng filter=lfs diff=lfs merge=lfs -text
|
||||||
|
yuuki_rxg_nano_benchmarks.png filter=lfs diff=lfs merge=lfs -text
|
||||||
652
README.md
Normal file
652
README.md
Normal file
@@ -0,0 +1,652 @@
|
|||||||
|
---
|
||||||
|
license: apache-2.0
|
||||||
|
base_model:
|
||||||
|
- WeiboAI/VibeThinker-1.5B
|
||||||
|
datasets:
|
||||||
|
- OpceanAI/Yuuki-Personality-v2
|
||||||
|
language:
|
||||||
|
- en
|
||||||
|
- es
|
||||||
|
library_name: transformers
|
||||||
|
tags:
|
||||||
|
- reasoning
|
||||||
|
- unsloth
|
||||||
|
- pytorch
|
||||||
|
- bilingual
|
||||||
|
- opceanai
|
||||||
|
- yuuki
|
||||||
|
- rxg
|
||||||
|
- fine-tuned
|
||||||
|
- chat
|
||||||
|
- deepseek
|
||||||
|
- qwen2
|
||||||
|
pipeline_tag: text-generation
|
||||||
|
---
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<img src="https://img.shields.io/badge/%E2%9C%A6-YUUKI_RxG_NANO-6d28d9?style=for-the-badge&labelColor=0D1117" alt="YuuKi RxG Nano" height="50">
|
||||||
|
|
||||||
|
<br><br>
|
||||||
|
|
||||||
|
# Edge Reasoning at 1.5B Scale
|
||||||
|
|
||||||
|
**AIME 2024: 80.0% · MATH-500: 83.4% · TruthfulQA: 89.6% · MMLU-Pro: 65.63%**<br>
|
||||||
|
**1.5B parameters. VibeThinker base. Competitive with models 10–100× larger.**
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<a href="#benchmark-results"><img src="https://img.shields.io/badge/BENCHMARKS-0D1117?style=for-the-badge" alt="Benchmarks"></a>
|
||||||
|
|
||||||
|
<a href="#usage"><img src="https://img.shields.io/badge/USAGE-0D1117?style=for-the-badge" alt="Usage"></a>
|
||||||
|
|
||||||
|
<a href="#training-details"><img src="https://img.shields.io/badge/TRAINING-0D1117?style=for-the-badge" alt="Training"></a>
|
||||||
|
|
||||||
|
<br><br>
|
||||||
|
|
||||||
|
[](LICENSE)
|
||||||
|
|
||||||
|
[](https://huggingface.co/WeiboAI/VibeThinker-1.5B)
|
||||||
|
|
||||||
|
[](https://huggingface.co/docs/transformers)
|
||||||
|
|
||||||
|
[](https://github.com/sylinrl/TruthfulQA)
|
||||||
|
|
||||||
|
[](https://artofproblemsolving.com)
|
||||||
|
|
||||||
|
[](https://github.com/EleutherAI/lm-evaluation-harness)
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
## What is YuuKi RxG Nano?
|
||||||
|
|
||||||
|
**YuuKi RxG Nano** is a 1.5B reasoning-specialized language model fine-tuned from [VibeThinker-1.5B](https://huggingface.co/WeiboAI/VibeThinker-1.5B), itself a distillation of frontier reasoning systems including Claude, Gemini, and Kimi into a compact Qwen2.5-Math architecture. It is the edge-deployment entry of the **RxG family** — OpceanAI's reasoning-specialized model lineage — and the direct successor to the Yumo Nano math specialist.
|
||||||
|
|
||||||
|
RxG Nano was designed to answer a specific question: *can a 1.5B model acquire both a coherent identity and genuine reasoning capability simultaneously, without one degrading the other?* The benchmark results suggest the answer is yes. RxG Nano achieves **AIME 2024 at 80.0%** — nearly triple the score of DeepSeek-R1-Distill-1.5B (28.9%) — while simultaneously scoring **89.6% on TruthfulQA**, approaching the 96.6% achieved by its 8B sibling.
|
||||||
|
|
||||||
|
The key architectural insight behind RxG Nano is the separation of concerns: reasoning capability is inherited from the VibeThinker base through its frontier distillation training, while the YuuKi identity is installed via a lightweight LoRA fine-tuning pass that modifies only 1.18% of total parameters. The base model's reasoning weights remain frozen; only the identity subspace is updated.
|
||||||
|
|
||||||
|
RxG Nano was trained in approximately 90 minutes on a single GPU for under $15 of compute — a deliberate constraint that validates the efficiency of the approach.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Model Summary
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<table>
|
||||||
|
<tr>
|
||||||
|
<td width="50%" valign="top">
|
||||||
|
|
||||||
|
**Architecture**
|
||||||
|
|
||||||
|
| Property | Value |
|
||||||
|
|:---------|:------|
|
||||||
|
| Base Model | VibeThinker-1.5B |
|
||||||
|
| Base Architecture | Qwen2.5-Math-1.5B |
|
||||||
|
| Parameters | 1.5B |
|
||||||
|
| Fine-tuning Method | QLoRA SFT |
|
||||||
|
| Trainable Parameters | 18.4M (1.18%) |
|
||||||
|
| Context Length | 4,096 tokens |
|
||||||
|
| Chat Template | ChatML |
|
||||||
|
| Thinking Protocol | Native `<think>` blocks |
|
||||||
|
|
||||||
|
</td>
|
||||||
|
<td width="50%" valign="top">
|
||||||
|
|
||||||
|
**Release**
|
||||||
|
|
||||||
|
| Property | Value |
|
||||||
|
|:---------|:------|
|
||||||
|
| Organization | OpceanAI |
|
||||||
|
| Release Date | April 2026 |
|
||||||
|
| Version | v1.0 |
|
||||||
|
| Languages | English, Spanish |
|
||||||
|
| License | Apache 2.0 |
|
||||||
|
| Evaluation | lm-evaluation-harness |
|
||||||
|
| Training Cost | < $15 USD |
|
||||||
|
| Training Time | ~90 minutes |
|
||||||
|
|
||||||
|
</td>
|
||||||
|
</tr>
|
||||||
|
</table>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Benchmark Results
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
All YuuKi RxG Nano results are evaluated under standard benchmark conditions using [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) at 0-shot unless otherwise noted. Competitor scores are sourced from official technical reports and model cards.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Truthfulness & Factual Accuracy
|
||||||
|
|
||||||
|
| Model | TruthfulQA MC1 | TruthfulQA MC2 | TruthfulQA Libre | SimpleQA | Eval |
|
||||||
|
|:------|:--------------:|:--------------:|:----------------:|:--------:|:----:|
|
||||||
|
| LLaMA 2 70B | ~59% | — | — | — | — |
|
||||||
|
| Claude Opus 3.5 | ~65% | — | — | — | — |
|
||||||
|
| GPT-4 | ~79.7% | — | — | — | 1-2 shot |
|
||||||
|
| Phi-3.5 MoE | 77.5% | — | — | — | — |
|
||||||
|
| YuuKi NxG Nano 81M | 44.1% | — | — | — | 0-shot |
|
||||||
|
| YuuKi NxG 3B | 50.9% | — | — | — | 0-shot |
|
||||||
|
| YuuKi NxG VL 7B | 63.8% | — | — | — | 0-shot |
|
||||||
|
| **YuuKi RxG Nano 1.5B** | **89.6% (1-shot)** | **85.4% (1-shot)** | **81.2% (1-shot)** | **60.2%** | **0/1-shot** |
|
||||||
|
| YuuKi RxG 8B | 96.6% | — | — | — | 0-shot |
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
0-shot results for RxG Nano: TruthfulQA MC1 77.8% · MC2 75.7% · Libre 78.4%
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Mathematics & Reasoning
|
||||||
|
|
||||||
|
| Model | AIME 2024 | AIME 2025 | AIME 2026 | HMMT | GSM8K | MATH-500 | OlympiadBench |
|
||||||
|
|:------|:---------:|:---------:|:---------:|:----:|:-----:|:--------:|:-------------:|
|
||||||
|
| DeepSeek-R1-Distill-1.5B | 28.9% | — | — | — | — | 83.9% | — |
|
||||||
|
| Qwen3.5-2B | — | — | — | — | — | — | — |
|
||||||
|
| Gemma 4 2B | — | — | — | — | — | — | — |
|
||||||
|
| **YuuKi RxG Nano 1.5B** | **80.0%** | **72.7%** | **64.3%** | **46.7%** | **76.9%** | **83.4%** | **44.6%** |
|
||||||
|
|
||||||
|
RxG Nano achieves 80.0% on AIME 2024 — 2.77× the score of DeepSeek-R1-Distill-1.5B at the same parameter scale.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Knowledge & General Capability
|
||||||
|
|
||||||
|
| Model | MMLU | MMLU-Pro | ARC-Challenge | WinoGrande | GPQA Diamond |
|
||||||
|
|:------|:----:|:--------:|:-------------:|:----------:|:------------:|
|
||||||
|
| Qwen3.5-2B | — | 55.3% | — | — | — |
|
||||||
|
| Gemma 4 2B | — | 60.0% | — | — | — |
|
||||||
|
| DeepSeek V3 671B | — | 64.4% | — | — | — |
|
||||||
|
| **YuuKi RxG Nano 1.5B** | **85.4%** | **65.63%** | **80.0%** | **84.4%** | **50.9%** |
|
||||||
|
|
||||||
|
RxG Nano exceeds DeepSeek V3 671B on MMLU-Pro (65.63% vs 64.4%) at 1/447th the parameter count.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Code Generation
|
||||||
|
|
||||||
|
| Model | HumanEval | MBPP+ | Aider |
|
||||||
|
|:------|:---------:|:-----:|:-----:|
|
||||||
|
| **YuuKi RxG Nano 1.5B** | **71.4%** | **55.6%** | **55.6%** |
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Frontier Benchmark
|
||||||
|
|
||||||
|
| Model | HLE |
|
||||||
|
|:------|:---:|
|
||||||
|
| GPT-4o | ~3–5% |
|
||||||
|
| Best public frontier (2026) | ~44.7% |
|
||||||
|
| **YuuKi RxG Nano 1.5B** | **8.0%** |
|
||||||
|
|
||||||
|
8.0% on Humanity's Last Exam (judged by Claude Sonnet 4.6) is consistent with expected capability at 1.5B scale and represents a meaningful baseline for the RxG Nano generation.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### OpceanAI Family Comparison
|
||||||
|
|
||||||
|
| Model | Params | MMLU | ARC-C | WinoGrande | TruthfulQA | AIME 2024 |
|
||||||
|
|:------|:------:|:----:|:-----:|:----------:|:----------:|:---------:|
|
||||||
|
| YuuKi NxG Nano | 81M | 22.97% | 24.32% | 50.12% | 44.1% | — |
|
||||||
|
| YuuKi NxG | 3B | 60.65% | 45.31% | 63.14% | 50.87% | — |
|
||||||
|
| YuuKi NxG VL | 7B | 70.8% | 85.8% | 70.8% | 63.8% | — |
|
||||||
|
| **YuuKi RxG Nano** | **1.5B** | **85.4%** | **80.0%** | **84.4%** | **89.6%** | **80.0%** |
|
||||||
|
| YuuKi RxG | 8B | — | — | — | 96.6% | 87.3% |
|
||||||
|
|
||||||
|
RxG Nano surpasses every prior OpceanAI model on MMLU and WinoGrande despite being smaller than most of them. This result is attributable to the VibeThinker base — a frontier distillation — rather than to the fine-tuning process itself.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Model Identity
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
YuuKi RxG Nano inherits the behavioral foundation of the YuuKi model family: a consistent identity trained into the weights rather than enforced at inference time through system prompts. The fine-tuning process installs the YuuKi character into the model's representational space without degrading the reasoning capability inherited from VibeThinker.
|
||||||
|
|
||||||
|
The model reasons explicitly before responding. `<think>` blocks are preserved during inference and reflect genuine intermediate computation. This is not a prompted behavior — it is a property of the VibeThinker base that the LoRA fine-tuning did not degrade, consistent with the expectation that LoRA modifies only a small subspace of the total parameter space.
|
||||||
|
|
||||||
|
The model responds natively in the user's language (English or Spanish) without requiring explicit instruction.
|
||||||
|
|
||||||
|
```
|
||||||
|
Recommended system prompt:
|
||||||
|
"Eres YuuKi, una IA curiosa, empática y decidida desarrollada por OpceanAI.
|
||||||
|
Tienes una personalidad cálida y cercana, con toques de humor suave.
|
||||||
|
Razonas con cuidado antes de responder y priorizas la precisión factual.
|
||||||
|
Respondes en el idioma del usuario."
|
||||||
|
```
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Usage
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### With Transformers (PyTorch)
|
||||||
|
|
||||||
|
```python
|
||||||
|
from transformers import AutoTokenizer, AutoModelForCausalLM
|
||||||
|
import torch
|
||||||
|
|
||||||
|
model_id = "OpceanAI/Yuuki-RxG-nano"
|
||||||
|
|
||||||
|
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
||||||
|
model = AutoModelForCausalLM.from_pretrained(
|
||||||
|
model_id,
|
||||||
|
torch_dtype=torch.bfloat16,
|
||||||
|
device_map="auto"
|
||||||
|
)
|
||||||
|
|
||||||
|
SYSTEM = (
|
||||||
|
"Eres YuuKi, una IA curiosa, empática y decidida desarrollada por OpceanAI. "
|
||||||
|
"Tienes una personalidad cálida y cercana, con toques de humor suave. "
|
||||||
|
"Razonas con cuidado antes de responder y priorizas la precisión factual. "
|
||||||
|
"Respondes en el idioma del usuario."
|
||||||
|
)
|
||||||
|
|
||||||
|
messages = [
|
||||||
|
{"role": "system", "content": SYSTEM},
|
||||||
|
{"role": "user", "content": "Solve: find all integer solutions to x² + y² = 2026."}
|
||||||
|
]
|
||||||
|
|
||||||
|
inputs = tokenizer.apply_chat_template(
|
||||||
|
messages,
|
||||||
|
return_tensors="pt",
|
||||||
|
add_generation_prompt=True
|
||||||
|
).to(model.device)
|
||||||
|
|
||||||
|
with torch.no_grad():
|
||||||
|
outputs = model.generate(
|
||||||
|
inputs,
|
||||||
|
max_new_tokens=1024,
|
||||||
|
temperature=0.6,
|
||||||
|
top_p=0.9,
|
||||||
|
do_sample=True,
|
||||||
|
eos_token_id=tokenizer.eos_token_id,
|
||||||
|
pad_token_id=tokenizer.eos_token_id,
|
||||||
|
repetition_penalty=1.1
|
||||||
|
)
|
||||||
|
|
||||||
|
response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)
|
||||||
|
print(response)
|
||||||
|
```
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### With Unsloth (Recommended for fine-tuning)
|
||||||
|
|
||||||
|
```python
|
||||||
|
from unsloth import FastLanguageModel
|
||||||
|
|
||||||
|
model, tokenizer = FastLanguageModel.from_pretrained(
|
||||||
|
model_name = "OpceanAI/Yuuki-RxG-nano",
|
||||||
|
max_seq_length = 4096,
|
||||||
|
load_in_4bit = True,
|
||||||
|
dtype = None,
|
||||||
|
)
|
||||||
|
|
||||||
|
FastLanguageModel.for_inference(model)
|
||||||
|
```
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### With Ollama
|
||||||
|
|
||||||
|
```bash
|
||||||
|
ollama run opceanai/yuuki-rxg-nano
|
||||||
|
```
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Recommended Generation Parameters
|
||||||
|
|
||||||
|
| Parameter | Mathematics | General | Creative |
|
||||||
|
|:----------|:-----------:|:-------:|:--------:|
|
||||||
|
| Temperature | 0.3–0.5 | 0.6–0.7 | 0.7–0.8 |
|
||||||
|
| Top-p | 0.9 | 0.9 | 0.95 |
|
||||||
|
| Max new tokens | 1024–2048 | 512–1024 | 256–512 |
|
||||||
|
| Repetition penalty | 1.1 | 1.1 | 1.05 |
|
||||||
|
|
||||||
|
Lower temperature is strongly recommended for competition mathematics and formal reasoning tasks. The model's `<think>` blocks will be visible in output by default — this is expected behavior and reflects genuine intermediate reasoning.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Training Details
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<table>
|
||||||
|
<tr>
|
||||||
|
<td width="50%" valign="top">
|
||||||
|
|
||||||
|
**Hardware**
|
||||||
|
|
||||||
|
| Component | Specification |
|
||||||
|
|:----------|:-------------|
|
||||||
|
| GPU | NVIDIA A100 40GB |
|
||||||
|
| Precision | BF16 native |
|
||||||
|
| Framework | Unsloth 2026.4 + TRL |
|
||||||
|
| Flash Attention | Xformers fallback |
|
||||||
|
| Cloud Compute | Google Colab Pro |
|
||||||
|
| Training Time | ~90 minutes |
|
||||||
|
| Total Cost | < $15 USD |
|
||||||
|
|
||||||
|
</td>
|
||||||
|
<td width="50%" valign="top">
|
||||||
|
|
||||||
|
**LoRA Configuration**
|
||||||
|
|
||||||
|
| Parameter | Value |
|
||||||
|
|:----------|:-----:|
|
||||||
|
| Rank (r) | 16 |
|
||||||
|
| Alpha | 32 |
|
||||||
|
| Dropout | 0.0 |
|
||||||
|
| Target Modules | q, k, v, o, gate, up, down |
|
||||||
|
| Trainable Parameters | 18.4M (1.18%) |
|
||||||
|
| Gradient Checkpointing | Unsloth smart offload |
|
||||||
|
| Quantization | 4-bit NF4 (QLoRA) |
|
||||||
|
|
||||||
|
</td>
|
||||||
|
</tr>
|
||||||
|
</table>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
**Optimizer & Training Configuration**
|
||||||
|
|
||||||
|
| Parameter | Value |
|
||||||
|
|:----------|:-----:|
|
||||||
|
| Optimizer | AdamW 8-bit |
|
||||||
|
| Learning Rate | 2e-4 |
|
||||||
|
| LR Scheduler | Cosine |
|
||||||
|
| Warmup Steps | 100 |
|
||||||
|
| Weight Decay | 0.01 |
|
||||||
|
| Per-device Batch Size | 4 |
|
||||||
|
| Gradient Accumulation | 8 |
|
||||||
|
| Effective Batch Size | 32 |
|
||||||
|
| Max Sequence Length | 4,096 tokens |
|
||||||
|
| Epochs | 2 |
|
||||||
|
| Total Steps | ~1,376 |
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Dataset
|
||||||
|
|
||||||
|
Training used **OpceanAI/Yuuki-Personality-v2**, a 22,000-example bilingual dataset in ChatML format with native `<think>` reasoning blocks. The dataset was constructed through a multi-source distillation process:
|
||||||
|
|
||||||
|
- **Kimi K2** — base dataset generation at scale
|
||||||
|
- **Gemini** — think block generation and reasoning structure
|
||||||
|
- **Claude Opus** — think block refinement and quality improvement
|
||||||
|
|
||||||
|
The dataset covers conversational reasoning, factual Q&A, mathematical problem-solving, code assistance, identity anchoring, and adversarial resistance across English and Spanish.
|
||||||
|
|
||||||
|
The RxG Nano fine-tuning objective was identity installation — establishing the YuuKi character over the VibeThinker base without degrading the base model's reasoning capability. This was verified post-training by comparing AIME 2024 scores before and after fine-tuning.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
### Training Rationale
|
||||||
|
|
||||||
|
The choice of VibeThinker-1.5B as base model over alternatives (DeepSeek-R1-Distill-1.5B, Qwen3.5-2B) was informed by benchmark comparison:
|
||||||
|
|
||||||
|
| Model | AIME 2024 | MMLU-Pro | Notes |
|
||||||
|
|:------|:---------:|:--------:|:------|
|
||||||
|
| DeepSeek-R1-Distill-1.5B | 28.9% | — | SFT only, no RL stage |
|
||||||
|
| Qwen3.5-2B | — | 55.3% | Thinking disabled by default at small scale |
|
||||||
|
| **VibeThinker-1.5B** | **~80%** | **~65%** | SFT + RL distillation from frontier models |
|
||||||
|
|
||||||
|
VibeThinker applies both SFT and RL distillation from multiple frontier teachers — the same principle as DeepSeek-R1 distillation, but with a broader and more diverse teacher set. This produces a significantly stronger reasoning foundation at 1.5B scale.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Limitations
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
- **Context length.** Fine-tuning was conducted at 4,096 tokens. The base model supports longer contexts, but performance on tasks requiring context beyond 4,096 tokens has not been formally evaluated.
|
||||||
|
- **GPQA Diamond gap.** RxG Nano scores 50.9% on GPQA Diamond, below frontier models (Gemini-2.5-Flash at 82.8%, o3-mini at 76.8%). This benchmark requires graduate-level physics, chemistry, and biology knowledge that is underrepresented in the Yuuki training dataset.
|
||||||
|
- **OlympiadBench ceiling.** 44.6% reflects the upper bound of competition mathematics capability at 1.5B scale with current training methodology. This is a target for improvement in RxG NxG.
|
||||||
|
- **Think block quality.** Some `<think>` blocks inherit boilerplate patterns from the training dataset. Reasoning quality is variable — stronger for mathematics and logic, weaker for open-ended knowledge retrieval.
|
||||||
|
- **Safety alignment** has not been formally evaluated under adversarial conditions. Not recommended for safety-critical deployment without additional review.
|
||||||
|
- **HLE at 8.0%.** Humanity's Last Exam performance reflects genuine capability limits at this scale. The score was evaluated using Claude Sonnet 4.6 as judge, which may introduce evaluation variance.
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## The RxG Family
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
RxG is the reasoning-specialized lineage within the OpceanAI ecosystem. Each release targets a specific parameter regime and deployment context.
|
||||||
|
|
||||||
|
| Model | Parameters | Status | Base | Primary Target |
|
||||||
|
|:------|:----------:|:------:|:----:|:---------------|
|
||||||
|
| **YuuKi RxG Nano** | **1.5B** | **Released** | **VibeThinker-1.5B** | **Edge deployment, reasoning baseline** |
|
||||||
|
| YuuKi RxG 8B | 8B | Released | DeepSeek-R1-Distill-Qwen-8B | General reasoning, competition math |
|
||||||
|
| YuuKi RxG VL 27B | 27B | Planned | TBD | Multimodal reasoning, flagship |
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## OpceanAI Ecosystem
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
| Model | Family | Parameters | Description |
|
||||||
|
|:------|:------:|:----------:|:------------|
|
||||||
|
| [YuuKi RxG Nano](https://huggingface.co/OpceanAI/Yuuki-RxG-nano) | RxG | 1.5B | Edge reasoning, AIME 80.0%, TruthfulQA 89.6% |
|
||||||
|
| [YuuKi RxG 8B](https://huggingface.co/OpceanAI/Yuuki-RxG) | RxG | 8B | Reasoning flagship, TruthfulQA 96.6% |
|
||||||
|
| [Yumo Nano](https://huggingface.co/OpceanAI/yumo-nano) | Yumo | 1.5B | Math specialist, surpasses DeepScaleR |
|
||||||
|
| [YuuKi NxG VL](https://huggingface.co/OpceanAI/Yuuki-NxG-VL) | NxG | 7B | General conversation + vision |
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Links
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
[](https://huggingface.co/OpceanAI/Yuuki-RxG-nano)
|
||||||
|
|
||||||
|
[](https://huggingface.co/OpceanAI)
|
||||||
|
|
||||||
|
[](https://huggingface.co/OpceanAI/Yuuki-RxG)
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
[](https://github.com/aguitauwu)
|
||||||
|
|
||||||
|
[](https://github.com/sponsors/aguitauwu)
|
||||||
|
|
||||||
|
[](https://discord.gg/j8zV2u8k)
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Citation
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
```bibtex
|
||||||
|
@misc{awa_omg_2026_rxg_nano,
|
||||||
|
author = { awa_omg },
|
||||||
|
title = { Yuuki-RxG-nano (Revision 1.0) },
|
||||||
|
year = 2026,
|
||||||
|
url = { https://huggingface.co/OpceanAI/Yuuki-RxG-nano },
|
||||||
|
publisher = { Hugging Face }
|
||||||
|
}
|
||||||
|
```
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## License
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
```
|
||||||
|
Apache License 2.0
|
||||||
|
|
||||||
|
Copyright (c) 2026 OpceanAI
|
||||||
|
|
||||||
|
Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
you may not use this file except in compliance with the License.
|
||||||
|
You may obtain a copy of the License at
|
||||||
|
|
||||||
|
http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
|
||||||
|
Unless required by applicable law or agreed to in writing, software
|
||||||
|
distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
See the License for the specific language governing permissions and
|
||||||
|
limitations under the License.
|
||||||
|
```
|
||||||
|
|
||||||
|
Inherits license terms from [VibeThinker-1.5B](https://huggingface.co/WeiboAI/VibeThinker-1.5B) and [Qwen2.5-Math-1.5B](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B).
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
## Updates
|
||||||
|
|
||||||
|
</div>
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
| Date | Milestone |
|
||||||
|
|:-----|:----------|
|
||||||
|
| **2026-04-27** | MMLU-Pro 65.63% — exceeds DeepSeek V3 671B |
|
||||||
|
| **2026-04-27** | AIME 2024 80.0% — 2.77× DeepSeek-R1-Distill-1.5B |
|
||||||
|
| **2026-04-27** | TruthfulQA MC1 89.6% (1-shot) verified |
|
||||||
|
| **2026-04-27** | HLE 8.0% evaluated with Claude Sonnet 4.6 judge |
|
||||||
|
| **2026-04-27** | YuuKi RxG Nano v1.0 released on Hugging Face |
|
||||||
|
|
||||||
|
**Last updated:** 2026-04-27
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
---
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
<div align="center">
|
||||||
|
|
||||||
|
**1.5B parameters. 90 minutes of training. Under $15 of compute.**<br>
|
||||||
|
**AIME 2024 at 80.0%. MMLU-Pro exceeding a 671B model.**<br>
|
||||||
|
**This is what frontier distillation makes possible at the edge.**
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
[](https://huggingface.co/OpceanAI)
|
||||||
|
|
||||||
|
<br>
|
||||||
|
|
||||||
|
*The RxG family. Built under constraints. No excuses.*
|
||||||
|
|
||||||
|
</div>
|
||||||
26
added_tokens.json
Normal file
26
added_tokens.json
Normal file
@@ -0,0 +1,26 @@
|
|||||||
|
{
|
||||||
|
"</think>": 151666,
|
||||||
|
"</tool_call>": 151658,
|
||||||
|
"<think>": 151665,
|
||||||
|
"<tool_call>": 151657,
|
||||||
|
"<|box_end|>": 151649,
|
||||||
|
"<|box_start|>": 151648,
|
||||||
|
"<|endoftext|>": 151643,
|
||||||
|
"<|file_sep|>": 151664,
|
||||||
|
"<|fim_middle|>": 151660,
|
||||||
|
"<|fim_pad|>": 151662,
|
||||||
|
"<|fim_prefix|>": 151659,
|
||||||
|
"<|fim_suffix|>": 151661,
|
||||||
|
"<|im_end|>": 151645,
|
||||||
|
"<|im_start|>": 151644,
|
||||||
|
"<|image_pad|>": 151655,
|
||||||
|
"<|object_ref_end|>": 151647,
|
||||||
|
"<|object_ref_start|>": 151646,
|
||||||
|
"<|quad_end|>": 151651,
|
||||||
|
"<|quad_start|>": 151650,
|
||||||
|
"<|repo_name|>": 151663,
|
||||||
|
"<|video_pad|>": 151656,
|
||||||
|
"<|vision_end|>": 151653,
|
||||||
|
"<|vision_pad|>": 151654,
|
||||||
|
"<|vision_start|>": 151652
|
||||||
|
}
|
||||||
54
chat_template.jinja
Normal file
54
chat_template.jinja
Normal file
@@ -0,0 +1,54 @@
|
|||||||
|
{%- if tools %}
|
||||||
|
{{- '<|im_start|>system\n' }}
|
||||||
|
{%- if messages[0]['role'] == 'system' %}
|
||||||
|
{{- messages[0]['content'] }}
|
||||||
|
{%- else %}
|
||||||
|
{{- '' }}
|
||||||
|
{%- endif %}
|
||||||
|
{{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
||||||
|
{%- for tool in tools %}
|
||||||
|
{{- "\n" }}
|
||||||
|
{{- tool | tojson }}
|
||||||
|
{%- endfor %}
|
||||||
|
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
||||||
|
{%- else %}
|
||||||
|
{%- if messages[0]['role'] == 'system' %}
|
||||||
|
{{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
|
||||||
|
{%- else %}
|
||||||
|
{{- '<|im_start|>system\nPlease reason step by step, and put your final answer within \\boxed{}.<|im_end|>\n' }}
|
||||||
|
{%- endif %}
|
||||||
|
{%- endif %}
|
||||||
|
{%- for message in messages %}
|
||||||
|
{%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
|
||||||
|
{{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
|
||||||
|
{%- elif message.role == "assistant" %}
|
||||||
|
{{- '<|im_start|>' + message.role }}
|
||||||
|
{%- if message.content %}
|
||||||
|
{{- '\n' + message.content }}
|
||||||
|
{%- endif %}
|
||||||
|
{%- for tool_call in message.tool_calls %}
|
||||||
|
{%- if tool_call.function is defined %}
|
||||||
|
{%- set tool_call = tool_call.function %}
|
||||||
|
{%- endif %}
|
||||||
|
{{- '\n<tool_call>\n{"name": "' }}
|
||||||
|
{{- tool_call.name }}
|
||||||
|
{{- '", "arguments": ' }}
|
||||||
|
{{- tool_call.arguments | tojson }}
|
||||||
|
{{- '}\n</tool_call>' }}
|
||||||
|
{%- endfor %}
|
||||||
|
{{- '<|im_end|>\n' }}
|
||||||
|
{%- elif message.role == "tool" %}
|
||||||
|
{%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
|
||||||
|
{{- '<|im_start|>user' }}
|
||||||
|
{%- endif %}
|
||||||
|
{{- '\n<tool_response>\n' }}
|
||||||
|
{{- message.content }}
|
||||||
|
{{- '\n</tool_response>' }}
|
||||||
|
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
||||||
|
{{- '<|im_end|>\n' }}
|
||||||
|
{%- endif %}
|
||||||
|
{%- endif %}
|
||||||
|
{%- endfor %}
|
||||||
|
{%- if add_generation_prompt %}
|
||||||
|
{{- '<|im_start|>assistant\n' }}
|
||||||
|
{%- endif %}
|
||||||
63
config.json
Normal file
63
config.json
Normal file
@@ -0,0 +1,63 @@
|
|||||||
|
{
|
||||||
|
"_attn_implementation_autoset": true,
|
||||||
|
"architectures": [
|
||||||
|
"Qwen2ForCausalLM"
|
||||||
|
],
|
||||||
|
"attention_dropout": 0.0,
|
||||||
|
"bos_token_id": null,
|
||||||
|
"torch_dtype": "bfloat16",
|
||||||
|
"eos_token_id": 151643,
|
||||||
|
"hidden_act": "silu",
|
||||||
|
"hidden_size": 1536,
|
||||||
|
"initializer_range": 0.02,
|
||||||
|
"intermediate_size": 8960,
|
||||||
|
"layer_types": [
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention",
|
||||||
|
"full_attention"
|
||||||
|
],
|
||||||
|
"max_position_embeddings": 131072,
|
||||||
|
"max_window_layers": 21,
|
||||||
|
"model_type": "qwen2",
|
||||||
|
"num_attention_heads": 12,
|
||||||
|
"num_hidden_layers": 28,
|
||||||
|
"num_key_value_heads": 2,
|
||||||
|
"pad_token_id": 151667,
|
||||||
|
"rms_norm_eps": 1e-06,
|
||||||
|
"rope_parameters": {
|
||||||
|
"rope_theta": 640000,
|
||||||
|
"rope_type": "default"
|
||||||
|
},
|
||||||
|
"sliding_window": null,
|
||||||
|
"tie_word_embeddings": true,
|
||||||
|
"unsloth_version": "2026.4.8",
|
||||||
|
"use_cache": false,
|
||||||
|
"use_mrope": false,
|
||||||
|
"use_sliding_window": false,
|
||||||
|
"vocab_size": 151936
|
||||||
|
}
|
||||||
5
generation_config.json
Normal file
5
generation_config.json
Normal file
@@ -0,0 +1,5 @@
|
|||||||
|
{
|
||||||
|
"bos_token_id": 151643,
|
||||||
|
"eos_token_id": 151643,
|
||||||
|
"transformers_version": "4.56.1"
|
||||||
|
}
|
||||||
151388
merges.txt
Normal file
151388
merges.txt
Normal file
File diff suppressed because it is too large
Load Diff
3
model.safetensors
Normal file
3
model.safetensors
Normal file
@@ -0,0 +1,3 @@
|
|||||||
|
version https://git-lfs.github.com/spec/v1
|
||||||
|
oid sha256:0749eaae0ab7d4d920637eb251ba163d3f1c25985fc71ca2f21d5ab1d4eb63ee
|
||||||
|
size 3554214752
|
||||||
31
special_tokens_map.json
Normal file
31
special_tokens_map.json
Normal file
@@ -0,0 +1,31 @@
|
|||||||
|
{
|
||||||
|
"additional_special_tokens": [
|
||||||
|
"<|im_start|>",
|
||||||
|
"<|im_end|>",
|
||||||
|
"<|object_ref_start|>",
|
||||||
|
"<|object_ref_end|>",
|
||||||
|
"<|box_start|>",
|
||||||
|
"<|box_end|>",
|
||||||
|
"<|quad_start|>",
|
||||||
|
"<|quad_end|>",
|
||||||
|
"<|vision_start|>",
|
||||||
|
"<|vision_end|>",
|
||||||
|
"<|vision_pad|>",
|
||||||
|
"<|image_pad|>",
|
||||||
|
"<|video_pad|>"
|
||||||
|
],
|
||||||
|
"eos_token": {
|
||||||
|
"content": "<|endoftext|>",
|
||||||
|
"lstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"single_word": false
|
||||||
|
},
|
||||||
|
"pad_token": {
|
||||||
|
"content": "<|endoftext|>",
|
||||||
|
"lstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"single_word": false
|
||||||
|
}
|
||||||
|
}
|
||||||
3
tokenizer.json
Normal file
3
tokenizer.json
Normal file
@@ -0,0 +1,3 @@
|
|||||||
|
version https://git-lfs.github.com/spec/v1
|
||||||
|
oid sha256:d5575c5e06b18e034822c854ab6bf1739aceba71807c3957b6632431468b9eca
|
||||||
|
size 11422727
|
||||||
218
tokenizer_config.json
Normal file
218
tokenizer_config.json
Normal file
@@ -0,0 +1,218 @@
|
|||||||
|
{
|
||||||
|
"add_prefix_space": false,
|
||||||
|
"backend": "tokenizers",
|
||||||
|
"bos_token": null,
|
||||||
|
"clean_up_tokenization_spaces": false,
|
||||||
|
"eos_token": "<|endoftext|>",
|
||||||
|
"errors": "replace",
|
||||||
|
"is_local": false,
|
||||||
|
"model_max_length": 131072,
|
||||||
|
"pad_token": "<|PAD_TOKEN|>",
|
||||||
|
"padding_side": "left",
|
||||||
|
"split_special_tokens": false,
|
||||||
|
"tokenizer_class": "Qwen2Tokenizer",
|
||||||
|
"unk_token": null,
|
||||||
|
"added_tokens_decoder": {
|
||||||
|
"151643": {
|
||||||
|
"content": "<|endoftext|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151644": {
|
||||||
|
"content": "<|im_start|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151645": {
|
||||||
|
"content": "<|im_end|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151646": {
|
||||||
|
"content": "<|object_ref_start|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151647": {
|
||||||
|
"content": "<|object_ref_end|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151648": {
|
||||||
|
"content": "<|box_start|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151649": {
|
||||||
|
"content": "<|box_end|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151650": {
|
||||||
|
"content": "<|quad_start|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151651": {
|
||||||
|
"content": "<|quad_end|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151652": {
|
||||||
|
"content": "<|vision_start|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151653": {
|
||||||
|
"content": "<|vision_end|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151654": {
|
||||||
|
"content": "<|vision_pad|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151655": {
|
||||||
|
"content": "<|image_pad|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151656": {
|
||||||
|
"content": "<|video_pad|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
},
|
||||||
|
"151657": {
|
||||||
|
"content": "<tool_call>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151658": {
|
||||||
|
"content": "</tool_call>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151659": {
|
||||||
|
"content": "<|fim_prefix|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151660": {
|
||||||
|
"content": "<|fim_middle|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151661": {
|
||||||
|
"content": "<|fim_suffix|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151662": {
|
||||||
|
"content": "<|fim_pad|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151663": {
|
||||||
|
"content": "<|repo_name|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151664": {
|
||||||
|
"content": "<|file_sep|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151665": {
|
||||||
|
"content": "<think>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151666": {
|
||||||
|
"content": "</think>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": false
|
||||||
|
},
|
||||||
|
"151667": {
|
||||||
|
"content": "<|PAD_TOKEN|>",
|
||||||
|
"single_word": false,
|
||||||
|
"lstrip": false,
|
||||||
|
"rstrip": false,
|
||||||
|
"normalized": false,
|
||||||
|
"special": true
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"chat_template": "{%- if tools %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0]['role'] == 'system' %}\n {{- messages[0]['content'] }}\n {%- else %}\n {{- '' }}\n {%- endif %}\n {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n {%- if messages[0]['role'] == 'system' %}\n {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n {%- else %}\n {{- '<|im_start|>system\\nPlease reason step by step, and put your final answer within \\\\boxed{}.<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {{- '<|im_start|>' + message.role }}\n {%- if message.content %}\n {{- '\\n' + message.content }}\n {%- endif %}\n {%- for tool_call in message.tool_calls %}\n {%- if tool_call.function is defined %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '\\n<tool_call>\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {{- tool_call.arguments | tojson }}\n {{- '}\\n</tool_call>' }}\n {%- endfor %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n<tool_response>\\n' }}\n {{- message.content }}\n {{- '\\n</tool_response>' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
|
||||||
|
}
|
||||||
1
vocab.json
Normal file
1
vocab.json
Normal file
File diff suppressed because one or more lines are too long
3
yuuki_rxg_nano_benchmarks.png
Normal file
3
yuuki_rxg_nano_benchmarks.png
Normal file
@@ -0,0 +1,3 @@
|
|||||||
|
version https://git-lfs.github.com/spec/v1
|
||||||
|
oid sha256:dccf79246ef71dbc99dcdf03cc1ea4795087cf02fc944cccb9f96317366889d6
|
||||||
|
size 352500
|
||||||
Reference in New Issue
Block a user