Fine-tuned Qwen2.5-7B to 96% of Claude Haiku on a domain-specific task using ~$3 of API calls and zero human labelers
Summary
Presented DV-DPO, a method to fine-tune Qwen2.5-7B on domain-specific tasks using only ~$3 in API calls and zero human labelers, achieving 96% composite performance of Claude Haiku via adversarial cross-examination.
Similar Articles
@cniongolo: I’m not sure people realize yet that you can actually run Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated-MTP-GGUF on a dua…
Demonstrates running a custom Qwen model (Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated-MTP-GGUF) on dual Nvidia RTX PRO 6000 Blackwell GPUs at 195 tokens per second using Hugging Face Inference.
LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification
This paper presents an iterative imbalance-aware fine-tuning approach using Qwen3-8B with QLoRA for psychological defense mechanism classification, achieving a macro F1 of 0.3917 and ranking 4th out of 21 teams in the PsyDefDetect 2026 shared task.
Good results fine tuning a local LLM like Qwen 3:0.6B to categorize questions
A developer fine-tunes a small Qwen 3 0.6B model using the Unsloth framework to categorize household questions, achieving good results with only 850 training examples.
Qwen 3.6 27B on DeepSWE
Qwen 3.6 27B scored 2% on the DeepSWE benchmark, placing 18/20 above Haiku 4.5 and Minimax M2.7, highlighting the gap between local and leading-edge models.
@rasbt: Crazy model! It actually uses the old Qwen2.5-Coder-3B stack and got really great performance with their post-training …
A 3B parameter model using the Qwen2.5-Coder-3B stack achieves coding benchmark scores comparable to Claude Opus 4.5, with detailed post-training techniques including synthetic data, filtering, two-stage SFT, and a novel RL method (MGPO).