---
license: mit
language:
- en
- ru
library_name: transformers
pipeline_tag: text-generation
base_model: Qwen/Qwen3-0.6B-Base
tags:
- rag
- faithful-qa
- occ
---
# OCC-RAG-0.6B
GitHub |
Technical Report |
Cloud
**OCC-RAG-0.6B** is a 0.6B-parameter small language model specialized for **faithful, context-grounded question answering**. Along with OCC-RAG-1.7B, it belongs to the first generation of **Optimal Cognitive Core (OCC)** specialized reasoning models. Given a question and a set of sources, it produces a structured reasoning trace with explicit source citations, decides whether the context actually supports an answer, and either answers from the context or abstains.
Despite its size, OCC-RAG-0.6B matches or exceeds general-purpose models **2–6× larger** on multi-hop reasoning, faithfulness, and refusal benchmarks. It is mid-trained from `Qwen/Qwen3-0.6B-Base` on a large synthetic corpus of multi-context, multi-hop QA with citation-anchored reasoning traces.
## Highlights
- **Faithful by design** — answers only from the supplied context; achieves the best faithfulness (lowest memorization ratio) across all evaluated scales, including 32B models.
- **Calibrated abstention** — outputs `Not enough information` when the context does not support an answer.
- **Structured, citable reasoning** — every answer comes with a transparent trace (query analysis → source analysis → reasoning → status → answer) that cites sources by id.
- **Compact** — a small model that delivers chain-of-thought-level transparency at a fraction of full thinking-mode inference cost.
## Model overview
OCC-RAG-0.6B is mid-trained from `Qwen/Qwen3-0.6B-Base` via supervised fine-tuning on a synthetic corpus of **~3.25M QA pairs** (~2.78M single-hop, ~262k multi-hop single-context, ~165k multi-hop multi-context, and ~43k abstain examples), distilled from a larger teacher with citation-anchored reasoning traces. Multi-hop and multi-context subsets are oversampled to emphasize compositional reasoning. The prompt/response format is identical at training and inference time, so no train–test mismatch is introduced.
## Evaluation
Evaluated across multi-hop reasoning (HotpotQA, MuSiQue, TAT-QA), faithfulness (ConFiQA), and refusal (MuSiQue-Un). In-Acc = the gold answer appears as a substring of the prediction; F1 = token-level overlap between prediction and gold answer; M_R = memorization ratio (lower = more faithful); R-Acc = refusal accuracy.
| Model | HotpotQA
In-Acc | MuSiQue
In-Acc | TAT-QA
F1 | ConFiQA
In-Acc | ConFiQA
M_R ↓ | MuSiQue-Un
R-Acc |
|---|---|---|---|---|---|---|
| gemma-3-4b-it | 55.8 | 30.1 | 65.3 | 69.8 | 8.9 | 55.8 |
| Qwen3-1.7B (think) | 60.9 | 30.7 | 74.8 | 70.4 | 8.3 | 82.8 |
| Qwen3-4B (think) | 67.1 | 41.5 | 79.1 | 74.1 | 7.5 | 84.0 |
| Pleias-RAG-1.2B | 48.5 | 15.0 | 8.4 | 37.3 | 25.3 | 21.9 |
| **OCC-RAG-0.6B** | **57.6** | **36.6** | **75.0** | **79.9** | **5.2** | **86.9** |
OCC-RAG-0.6B exceeds Gemma-3-4B and SmolLM-3-3B on every dimension and attains the strongest faithfulness (highest ConFiQA In-Acc, lowest M_R) among all evaluated models.
## Input / output format
OCC-RAG uses a **structured prompt format with special tokens**. The question is wrapped in `<|query_start|> … <|query_end|>` and each source in `<|source_start|><|source_id|>N … <|source_end|>`.
The response is split into five sections, each delimited by special tokens:
| Section | Tokens | Content |
|---|---|---|
| Query analysis | `<\|query_analysis_start\|> … <\|query_analysis_end\|>` | Decomposes the question into what must be found. |
| Source analysis | `<\|source_analysis_start\|> … <\|source_analysis_end\|>` | Assesses each source's relevance, citing by `<\|source_id\|>N`. |
| Reasoning | `<\|reasoning_start\|> … <\|reasoning_end\|>` | Composes evidence across sources into a multi-hop chain. |
| Status | `<\|status_start\|> … <\|status_end\|>` | `ANSWERABLE` / `UNANSWERABLE` verdict. |
| Answer | `<\|answer_start\|> … <\|answer_end\|>` | The final answer span, or the refusal phrase. |
## Quickstart (Transformers)
The chat template accepts a `documents=` kwarg and emits the structural tokens for the query and sources automatically — pass the user message as plain text and the sources as a list of dicts.
```python
import re
from transformers import AutoModelForCausalLM, AutoTokenizer
MODEL = "occ-ai/OCC-RAG-0.6B"
tokenizer = AutoTokenizer.from_pretrained(MODEL)
model = AutoModelForCausalLM.from_pretrained(MODEL, torch_dtype="auto", device_map="auto")
question = "Which country is the inventor of the telephone, Alexander Graham Bell, buried in?"
documents = [
{"text": "Alexander Graham Bell was a Scottish-born inventor best known for patenting the first practical telephone."},
{"text": "Bell died on August 2, 1922, at his estate Beinn Bhreagh, near Baddeck, Nova Scotia, and was buried there."},
{"text": "Nova Scotia is a province on the east coast of Canada."},
]
text = tokenizer.apply_chat_template(
[{"role": "user", "content": question}],
documents=documents,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False,
)
# Alternative: assemble the structural tokens yourself.
#
# query_start, query_end = "<|query_start|>", "<|query_end|>"
# source_start, source_end, source_id = "<|source_start|>", "<|source_end|>", "<|source_id|>"
#
# def build_user_content(question, sources):
# content = f"{query_start}{question}{query_end}\n"
# for i, s in enumerate(sources, start=1):
# content += f"{source_start}{source_id}{i} {s}{source_end}\n"
# return content
#
# messages = [{"role": "user", "content": build_user_content(question, [d["text"] for d in documents])}]
# text = tokenizer.apply_chat_template(
# messages, tokenize=False, add_generation_prompt=True, enable_thinking=False
# )
inputs = tokenizer([text], return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=2048)
response = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=False)
print(response)
m = re.findall(r"<\|answer_start\|>(.*?)(?:<\|answer_end\|>|\Z)", response, re.DOTALL)
print("Answer:", m[-1].strip() if m else "") # -> Canada
```
> [!NOTE]
> We recommend greedy decoding (`do_sample=False`), which is the training/evaluation default and is baked into `generation_config.json`. Qwen3's default sampling parameters ([best practices](https://huggingface.co/Qwen/Qwen3-0.6B#best-practices)) also work fine.
## Deployment
OCC-RAG-0.6B is a standard Qwen3 causal LM and is compatible with vLLM, SGLang, and other Transformers-based serving stacks. With only 0.6B parameters, it can be readily deployed in constrained infrastructure, including desktop systems running on CPU RAM. When serving, keep `skip_special_tokens=False` if you need to parse the structural tokens out of the raw output.
When using an OpenAI-compatible server (vLLM ≥0.6, SGLang ≥0.4.7), the `documents=` kwarg is reachable from the client via `chat_template_kwargs`:
```python
client.chat.completions.create(
model="occ-ai/OCC-RAG-0.6B",
messages=[{"role": "user", "content": question}],
extra_body={"chat_template_kwargs": {"documents": documents}},
)
```
## Limitations
- **Context-grounded only.** The model is trained to answer from the supplied sources and to ignore parametric knowledge. It is not a general-purpose chat or knowledge model.
- **Reasoning depth.** Training and evaluation are capped at three-hop reasoning; longer chains are out of distribution.
## Citation
If you find our work helpful, feel free to give us a cite.
```bibtex
@misc{savkin2026occragoptimalcognitivecore,
title = {OCC-RAG: Optimal Cognitive Core for Faithful Question Answering},
author = {Maksim Savkin and Mikhail Goncharov and Alexander Gambashidze and Alla Chepurova and Dmitrii Tarasov and Nikita Andriianov and Daria Pugacheva and Vasily Konovalov and Andrey Galichin and Ivan Oseledets},
year = {2026},
eprint = {2606.00683},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
url = {https://arxiv.org/abs/2606.00683}
}
```