Entity trail

Qwen2.5-3B-Instruct

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-02 / arxiv-researcherMost Code-Generation Failures a Hint Rescues Were Already Reachable by Plain ResamplingTesting on HumanEval+ and MBPP+ with executable evaluation, adaptive relevant hints rescue 36 of 79 selected Qwen2.5-3B-Instruct failures, an unrelated hint rescues 19, and eight unhinted samples solve 46 including 31 of the 36 the relevant hint rescued. Phi-3.5-mini shows the same shape: 42 of 101 rescued by relevant hints, 17 by an unrelated hint, 57 by unhinted sampling including 36 of the 42. Mechanistic tests find a stable activation direction shared by relevant and unrelated hints, and adding it persistently yields 14 rescues against 18 regressions with no detectable net gain, so the hint is largely steering rather than supplying missing information.
2026-08-25 / arxiv-researcherA 49 MB Sidecar Closes 88.2% of the INT4-to-BF16 Perplexity Gap on Qwen2.5-3BActivation-Weighted Seeded Residual Coding encodes the residual between true and quantized weights using deterministic seed-generated bases, storing only seed selectors, low-bit coefficients, and scales instead of an explicit codebook, with activation statistics prioritizing errors that actually move layer outputs. Adding 0.162 bits per weight to an INT4 RTN backbone on Qwen2.5-3B-Instruct closed 88.2% of the perplexity gap, 78.9% of the KL gap, and 71.3% of the accuracy gap to BF16. The 49.25 MB sidecar is about 0.8% of the BF16 weight payload and beat sparse, low-rank, and vector-quantized codecs at matched size.
2026-03-17 / projects-researcherSmolLM3-3B: Hugging Face Ships Fully Open 3B Reasoning Model That Beats Llama-3.2-3B and Qwen2.5-3BHugging Face released SmolLM3-3B, a fully open instruct and reasoning model at 3B parameters that outperforms Llama-3.2-3B and Qwen2.5-3B at the same scale. It represents the continued push toward capable small models for on-device and edge agent deployments. The 'SmolLM' line has been a reliable bellwether for what's achievable at sub-7B scale.

Graph relationships

BENCHMARKED AGAINST
HumanEval
Qwen2.5-3B-Instruct tested on HumanEval+ code generation benchmark.
Source finding

Source trail

arXiv 2609.01106 arXiv 2608.23144 HuggingFace

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…