lewisdog/lfm2.5-350m-cogs-query-GGUF overview
LFM2.5 350M cogs query student GGUF LoRA fine tune of LiquidAI/LFM2.5 350M as the Cogitarium query student : translates a natural language data request + schem…
Runs locally from ~361.6 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
Model Details
Model README
---
license: other
license_name: lfm1.0
license_link: https://huggingface.co/LiquidAI/LFM2.5-350M/blob/main/LICENSE
base_model: LiquidAI/LFM2.5-350M
tags: [gguf, text-to-cypher, cypher, ladybugdb, cogitarium]
---
LFM2.5-350M cogs query student (GGUF)
LoRA fine-tune of LiquidAI/LFM2.5-350M as the Cogitarium query student:
translates a natural-language data request + schema into ONE LadybugDB Cypher
query (openCypher-like, constrained single-MATCH grammar). Trained on 3,065
execution-verified pairs (teacher glm-5.2; every pair executed against a
seeded store before admission).
Frozen 55-example execution eval (see the cogs-finetune-query runbook):
| model | exec | match |
|---|---|---|
| base LFM2.5-350M few-shot | 7.3% | 0% |
| Qwen3-1.7B few-shot | 94.5% | 12.7% |
| this model (bf16) | 98.2% | 23.6% |
| this model (Q8_0) | 96.4% | 23.6% |
| this model (Q4_K_M) | 63.6% | 10.9% |
Pins (measured): repetition penalty 1.0; Q8_0 quantization REQUIRED
(Q4_K_M exhibits token corruption). Serve with the exact system prompts in
Cogitarium src/data/prompts.ts (sha-pinned train/serve contract).
Run lewisdog/lfm2.5-350m-cogs-query-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models