fetch-llm-papers

name

fetch-llm-papers

description

Workflow for updating the LLM landscape paper pool (section/x_llm_papers.md) using fetch_llm_papers.py. Covers full re-fetch, resume from checkpoint, and adding new topics. USE FOR: Refreshing citation counts, expanding topic coverage. DO NOT USE FOR: Adding hand-curated entries to section files (use add-new-entry-from-temp-md), updating RAG/Agent citation sections in best_practices.md (use update-cite-count).

Overview

The pool file section/x_llm_papers.md is a compact list of high-citation CS papers covering the LLM landscape, fetched from the Semantic Scholar API and ranked by citation count. It includes a topic coverage summary and per-paper topic tags. It is generated and maintained by code/fetch_llm_papers.py.

The section ### **LLM Research (Ranked by cite count >=150)** in section/models_research.md links to this file with a single descriptive line.

---

Script Reference

Script: code/fetch_llm_papers.py
Python env: .venv\Scripts\python.exe

Key CLI Arguments

Argument	Default	Purpose
--output	section/x_llm_papers.md	Output markdown file for the paper pool
--min-citations	150	Minimum citation count filter
--top-n	50	Max papers returned per topic query
--request-delay	2.0	Delay between Semantic Scholar requests
--jitter	0.5	Random extra delay between requests
--api-key-env	S2_API_KEY	Env var containing a Semantic Scholar API key
--reset	(flag)	Delete existing checkpoint and start from scratch
--topics	(all)	Limit run to matching topic names (substring, case-insensitive)
--refresh-existing	(flag)	Refresh existing papers with the Semantic Scholar batch API, without search queries
--batch-size	100	Paper IDs per --refresh-existing batch
--annotate-existing	(flag)	Rewrite the existing markdown with inferred topic tags without API calls
--annotate-source	(output file)	Optional source for annotation; supports git:<rev>:<path>

---

Workflow

1. Refresh existing papers and citation counts

Use this for the fastest routine update when the current topic coverage is still appropriate.

.venv\Scripts\python.exe code/fetch_llm_papers.py `
    --refresh-existing `
    --min-citations 150 `
    --batch-size 100

• Uses Semantic Scholar's batch endpoint to update existing entries without re-running every search query.
• Preserves each paper's existing topic tags, removes papers that no longer meet the citation threshold, and rewrites the output in citation order.
• Use a lower --batch-size if unauthenticated requests are rate-limited.

2. Full re-fetch (refresh everything)

Use when topics have been added/modified or citation counts are stale.

.venv\Scripts\python.exe code/fetch_llm_papers.py `
    --reset `
    --min-citations 150 `
    --top-n 50 `
    --request-delay 2.0 `
    --jitter 0.5

• --reset deletes any existing checkpoint so all 41 topics are re-queried.
• On success the checkpoint is automatically deleted.
• section/x_llm_papers.md is rewritten with topic coverage, topic tags, and sequential numbering sorted by citation count.
• If a Semantic Scholar API key is available, set $env:S2_API_KEY before running. The script sends it as the x-api-key header.

3. Resume after API interruption

The script saves a checkpoint (section/x_llm_papers.checkpoint.json) after each successful query and after each completed topic. If the run is interrupted by a rate-limit (HTTP 429), simply re-run without --reset:

.venv\Scripts\python.exe code/fetch_llm_papers.py `
    --min-citations 150 `
    --top-n 50

The script prints [resume] Loaded N papers, M completed topics, Q cached queries from checkpoint. and skips already-finished topics. Query-level cache entries prevent successful queries from being reissued during resume.

If a query fails after retries, the script writes partial progress, keeps the current topic incomplete, and exits with a [pause] message. Re-run later without --reset.

4. Local annotation only (no API calls)

Use this when the paper pool already exists and you only need topic coverage or tag extraction refreshed:

.venv\Scripts\python.exe code/fetch_llm_papers.py `
    --annotate-existing `
    --min-citations 150

To rebuild annotations from the committed version of the file, useful after a partial write or parser change:

.venv\Scripts\python.exe code/fetch_llm_papers.py `
    --annotate-existing `
    --annotate-source git:HEAD:section/x_llm_papers.md `
    --min-citations 150

5. Refresh only specific topics

.venv\Scripts\python.exe code/fetch_llm_papers.py `
    --topics "PEFT" "Reasoning" `
    --min-citations 150 `
    --top-n 50

Matches topic names by substring (case-insensitive). New papers for matched topics are merged into the existing pool if a checkpoint exists; otherwise starts fresh for those topics only.

---

Adding or Modifying Topics

Topics are defined in the TOPICS dict at the top of fetch_llm_papers.py. Each key is a topic label; the value is a list of Semantic Scholar search query strings.

Rules:

• Queries should be descriptive phrases, not single keywords — Semantic Scholar full-text search works best with 4–8 word phrases.
• Avoid the word "survey" to capture research papers, benchmarks, and position papers, not just surveys.
• Aim for 3–10 queries per topic. Overlapping queries are fine — deduplication is handled automatically by paperId.
• After adding topics, run with --reset to re-fetch from scratch (checkpoint is stale once TOPICS changes).
• Keep broad queries paired with relevance terms such as LLM, language model, transformer, foundation model, agent, RAG, or tool use. The script filters broad Semantic Scholar drift using topic keywords plus core LLM relevance terms.

Current topic areas (41 total):

Category	Topics
Core LLM	Reasoning in LLMs, LLM Overview & History, Scaling Laws, LLM Architecture Innovations
Training	Alignment & RLHF, RLAIF & Constitutional AI, RLVR & Process Reward Models, Instruction Tuning & SFT, PEFT & LoRA, Self-Supervised & Representation Learning
Inference	Efficient LLMs: Training & Inference, Inference-Time Scaling & Test-Time Compute, LLMOps & Model Serving
Applications	LLM Agents, Retrieval-Augmented Generation (RAG), GraphRAG & Knowledge Graphs, LLMs for Code, LLMs for Healthcare & Science, LLM for Robotics & Embodied AI, Function Calling & Tool Use, GUI Agents, Tabular Data & NL2SQL
Multimodal	Multimodal LLMs, Multilingual & Low-Resource LLMs, Speech & Audio Language Models, Small Language Models, Mixture of Experts
Evaluation	Evaluation of LLMs & Agents, Hallucination in LLMs, Trustworthy & Secure LLMs
Generation	Structured Generation & Constrained Decoding
Other	Prompt Engineering & In-Context Learning, Context Engineering, LLM Memory & Personalization, Embeddings & Vector Search, Data for LLMs, LLM Governance, Privacy & Copyright, Interpretability & Mechanistic Understanding, AIOps & Observability, Federated & Personalized AI, Continual Learning & Model Merging

---

Output Format

Each entry in section/x_llm_papers.md:

N. [Title📑](https://arxiv.org/abs/XXXX.XXXXX): First sentence of abstract. [Mon YYYY] (Citations: N,NNN; Topics: Topic A, Topic B)

• The file begins with ## Topic Coverage, a count of inferred topic tags across all papers.
• Numbered sequentially (1., 2., ...) by citation count descending.
• Link target is the arXiv URL if available, otherwise the Semantic Scholar URL.
• Generated and retrieved timestamps are formatted in UTC. Paper dates are derived from the arXiv ID prefix (e.g. 2305.xxxxx → [May 2023]).
• Only Computer Science papers with fieldsOfStudy containing "Computer Science" are included.
• Topic tags come from the API topic that found the paper when fetched, or local keyword inference when running --annotate-existing.

---

Checkpoint File

section/x_llm_papers.checkpoint.json — JSON with these keys:

{
  "completed_topics": ["Reasoning in LLMs", "LLM Agents", ...],
    "failed_queries": ["..."],
    "query_cache": { "normalized query": [ ... ] },
  "papers": [ { "paperId": "...", "title": "...", "citationCount": 123, ... } ]
}

• Created/updated after each successful query and every topic completion.
• Deleted automatically on successful full run.
• If corrupted, delete manually and re-run with --reset.
• To inspect: python -c "import json; cp=json.load(open('section/x_llm_papers.checkpoint.json', encoding='utf-8')); print(len(cp['papers']), 'papers,', len(cp['completed_topics']), 'topics done,', len(cp.get('query_cache', {})), 'cached queries')"

---

Common Pitfalls

1. Modified TOPICS but not using --reset: The checkpoint from a previous run skips topics that already completed. After editing TOPICS, always use --reset to re-fetch all topics.

2. API rate limits (HTTP 429): Semantic Scholar enforces per-IP rate limits. The script adds a default 2-second delay plus jitter between queries and retries with backoff. If repeatedly rate-limited, wait a few minutes and resume (no --reset). Increase --request-delay or --backoff to slow down further.

3. Failed queries marked as complete: Do not manually add a topic to completed_topics after a 429 or timeout. The script intentionally leaves incomplete topics out of completed_topics so resume can retry only the missing query work.

4. Non-CS or off-topic papers in results: The filter requires "Computer Science" in fieldsOfStudy and a topic/core LLM relevance match. Some highly cited ML papers may be excluded if they drift too far from the LLM landscape. This is intentional.