Obsessive-1.7B: a 1.1 GB web-search subagent
A Reddit author fine-tuned LFM3-1.7B with LoRA rank 16 on about 9,000 tool-calling conversations to perform one task: answer a factual question with a short value, a word-for-word quote and a URL after calling web_search once, or say "Not found in the search results." The model was published as a 1.1 GB GGUF file on Hugging Face under the inherited LFM Open License, with configurations for Pi, OpenCode and Hermes and a llama-server command. In author-reported tests, it found answers in 50% of cases when available, invented 10% of answers and had a median response time of 5.1 seconds per question on an RTX 4060 laptop. In live Wikipedia search using an unseen backend, it returned a correct answer for 37 of 120 questions and answered one of 40 multi-fact requests. The author also reports that it says "not found" for about one in ten answers present in the results, supports only English and is suited to one question per conversation.
The release provides a small, runnable model specialized for one-shot web search beneath a larger model. Its author-reported evaluations show substantial limitations, including invented answers, missed answers and poor performance on multi-fact requests, so its usefulness depends on the deployment setup.
I fine-tuned LFM3-1.7B to do one job: take a factual question, call web_search once, and answer with a short value, a word-for-word quote and the URL. If the results don’t contain the answer, it says “Not found in the search results.” It’s meant to run as a subagent under a bigger model, not as a chat model. I have to say the base lfm really surprised me at its capability to run tool calls and extract information.
Answer: Barbara Liskov Evidence: "Liskov ... the second woman to receive the Turing Award." Source: https://en.wikipedia.org/wiki/Barbara_Liskov Fixed-results benchmark (131 held-out SimpleQA questions; in 29 of them the answer was removed from the results):
| base | Obsessive |
|---|---|
| Finds the answer when it’s there | 50% |
| Says “not found” when it isn’t | 0% |
| Invented answers | 10% |
| Median sec/question (4060 laptop) | 5.1 |
#Obsessive-1.7B#LFM3-1.7B#Liquid AI#llama.cpp#Hugging Face#Qwen 3.8 Flash Next