How Far Do On-Prem Open LLMs Get on Text-to-SQL? A Cross-Family Size x Technique Frontier on BIRD

• URL: http://arxiv.org/abs/2606.29733v1
• Date: Mon, 29 Jun 2026 03:15:38 GMT
• Title: How Far Do On-Prem Open LLMs Get on Text-to-SQL? A Cross-Family Size x Technique Frontier on BIRD
• Abstract summary: We evaluate three open model families across two generations -- Qwen2.5-Coder (7B/14B/32B), CodeLlama-Instruct (7B/13B/34B), and Llama-3.x (8B, 70B)<n>We find that Qwen2.5-Coder dominates the older CodeLlama at matched size (39.1 vs 20.9 at 7B), but a modern non-Qwen model (Llama-3.3-70B, 49.2 on a matched serving) is competitive, so CodeLlama's weakness reflects its 2023 generation, not "non-Qwen = weak"
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Organizations that cannot send data to a cloud API increasingly ask: how good is Text-to-SQL if the model must run on-premises on open weights, and which popular accuracy "recipes" are worth their compute? We answer with an honest, fully reproducible benchmark on the BIRD development split (n=1534, Execution Accuracy), evaluating three open model families across two generations -- Qwen2.5-Coder (7B/14B/32B), CodeLlama-Instruct (7B/13B/34B), and Llama-3.x (8B, 70B) -- under one matched protocol, ablating a model-agnostic recipe (schema linking, self-correction, self-consistency) component by component, with every difference tested by the paired McNemar test. Four findings stand out. (i) Generation matters more than raw size, and the recipe is family-robust: Qwen2.5-Coder dominates the older CodeLlama at matched size (39.1 vs 20.9 at 7B), but a modern non-Qwen model (Llama-3.3-70B, 49.2 on a matched serving) is competitive, so CodeLlama's weakness reflects its 2023 generation, not "non-Qwen = weak". (ii) Self-correction is a robust, near-free win, significant on all three families where there is room to improve. (iii) Schema linking does not help, and a stronger linker does not rescue it: a retrieval/embedding linker with 96.5% gold-table recall is statistically indistinguishable from no linking, ruling out the "weak lexical strawman" objection across three families. (iv) Self-consistency is poor value (+0.13 pp for ~5x tokens, not significant). We report real per-stage cost ($/1k queries) and release all code, predictions, and summaries; archived code and data: https://doi.org/10.5281/zenodo.20952794

Related papers

• Codebook Agent: Amortized Topology Design for LLM Multi-Agent Systems [67.04448659688579]
  We develop a query-independent 16-entry codebook that ranks the top decoded candidates in a single batched forward pass.<n>With no iterative search and no message passing at test time, Codebook Agent is the most accurate method on all six benchmarks.
  arXiv  Detail & Related papers  (2026-09-02T08:10:22Z)
• Beyond Pass Rate: A Multilingual, Execution-Grounded Evaluation of Open Code LLMs [0.0]
  Best model reaches 23.64% mean correctness, compared with a 57.2% human acceptance baseline.<n>Qwen2.5-Coder-14B-Instruct is strongest on hard problems and distinct-problem coverage.<n>Gemma-2-27B-IT achieves the highest all-language lint pass rate.
  arXiv  Detail & Related papers  (2026-06-07T21:10:30Z)
• Do not copy and paste! Rewriting strategies for code retrieval [0.7710436567988378]
  We study a hierarchy of three rewriting strategies: stylistic rephrasing, NL-enriched PseudoCode, and full Natural-Language transcription.<n>We are the first to evaluate NL-enriched PseudoCode and snippet-level Natural Language as direct retrieval representations, rather than as transient intermediates.<n>We introduce two diagnostics, Delta H, token entropy, and Delta s, embedding cosine, and show that Delta H predicts retrieval gain under QC across all three rewriter families.
  arXiv  Detail & Related papers  (2026-05-08T11:31:26Z)
• Semantic Layers for Reliable LLM-Powered Data Analytics: A Paired Benchmark of Accuracy and Hallucination Across Three Frontier Models [0.0]
  LLMs deployed for natural-language querying of analytical databases suffer from two intertwined failures.<n>We benchmark three frontier LLMs on 100 natural-language questions over the Cleaned Contoso Retail dataset in ClickHouse.
  arXiv  Detail & Related papers  (2026-04-28T02:53:23Z)
• vstash: Local-First Hybrid Retrieval with Adaptive Fusion for LLM Agents [0.0]
  We present **vstash**, a local-first document memory system that combines vector similarity search with fulltext keyword matching.<n>All data resides in a single file using sqlite-vec for approximate nearest neighbor search and FTS5 for keyword matching.
  arXiv  Detail & Related papers  (2026-04-16T19:22:58Z)
• Cross-Family Speculative Decoding for Polish Language Models on Apple~Silicon: An Empirical Evaluation of Bielik~11B with UAG-Extended MLX-LM [0.0]
  We extend the MLX-LM framework with Universal Assisted Generation (UAG) to enable cross-tokenizer speculative decoding on Apple Silicon.<n>Experiments on three Polish-language datasets (Wikipedia, pl_alpaca, synthetic) use draft lengths k in 2, 4, 6 to compare naive and context-aware token translation.
  arXiv  Detail & Related papers  (2026-03-22T19:07:41Z)
• ExCoT: Optimizing Reasoning for Text-to-SQL with Execution Feedback [49.21833666405111]
  Large language models (LLMs) excel in many reasoning tasks, but their ability to leverage Chain-of-Thought (CoT) reasoning remains underexplored.<n>We propose ExCoT, a novel framework that iteratively optimize open-source LLMs by combining CoT reasoning with off-policy and on-policy DPO.
  arXiv  Detail & Related papers  (2025-03-25T18:17:36Z)
• RSL-SQL: Robust Schema Linking in Text-to-SQL Generation [51.00761167842468]
  We propose a novel framework called RSL- that combines bidirectional schema linking, contextual information augmentation, binary selection strategy, and multi-turn self-correction. benchmarks demonstrate that our approach achieves SOTA execution accuracy among open-source solutions, with 67.2% on BIRD and 87.9% on GPT-4ocorrection. Our approach outperforms a series of GPT-4 based Text-to-Seek systems when adopting DeepSeek (much cheaper) with same intact prompts.
  arXiv  Detail & Related papers  (2024-10-31T16:22:26Z)
• LLM Robustness Against Misinformation in Biomedical Question Answering [50.98256373698759]
  The retrieval-augmented generation (RAG) approach is used to reduce the confabulation of large language models (LLMs) for question answering. We evaluate the effectiveness and robustness of four LLMs against misinformation in answering biomedical questions.
  arXiv  Detail & Related papers  (2024-10-27T16:23:26Z)
• Enhancing Self-Consistency and Performance of Pre-Trained Language Models through Natural Language Inference [72.61732440246954]
  Large pre-trained language models often lack logical consistency across test inputs. We propose a framework, ConCoRD, for boosting the consistency and accuracy of pre-trained NLP models. We show that ConCoRD consistently boosts accuracy and consistency of off-the-shelf closed-book QA and VQA models.
  arXiv  Detail & Related papers  (2022-11-21T21:58:30Z)
• DS-1000: A Natural and Reliable Benchmark for Data Science Code Generation [70.96868419971756]
  DS-1000 is a code generation benchmark with a thousand data science problems spanning seven Python libraries. First, our problems reflect diverse, realistic, and practical use cases since we collected them from StackOverflow. Second, our automatic evaluation is highly specific (reliable) -- across all Codex-predicted solutions that our evaluation accept, only 1.8% of them are incorrect.
  arXiv  Detail & Related papers  (2022-11-18T17:20:27Z)
• CoSQA: 20,000+ Web Queries for Code Search and Question Answering [63.92224685262063]
  CoSQA dataset includes 20,604 labels for pairs of natural language queries and codes. We introduce a contrastive learning method dubbed CoCLR to enhance query-code matching. We show that evaluated on CodeXGLUE with the same CodeBERT model, training on CoSQA improves the accuracy of code question answering by 5.1%.
  arXiv  Detail & Related papers  (2021-05-27T15:37:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.