Hugging Face's logo

Datasets:
LakoreAI
/
experiments

Tasks:
Question Answering
Languages:
English
Size:
n<1K
ArXiv:
Tags:
agent-memory
long-term-memory
rag
longmemeval
benchmark
blog-companion
License:
Dataset card Files
xet
 Community
minhleduc commited on
Commit
a0bfff9
·
verified ·
1 Parent(s): 26c4a5e

Add dataset card

Browse files
Files changed (1) hide show
  1. README.md +150 -0
README.md ADDED
@@ -0,0 +1,150 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ language:
4
+ - en
5
+ pretty_name: Blog Experiments (Minh Le Duc)
6
+ size_categories:
7
+ - n<1K
8
+ task_categories:
9
+ - question-answering
10
+ tags:
11
+ - agent-memory
12
+ - long-term-memory
13
+ - rag
14
+ - longmemeval
15
+ - benchmark
16
+ - blog-companion
17
+ viewer: false
18
+ ---
19
+
20
+ # Blog Experiments
21
+
22
+ Data files that sit behind the technical blog posts of **Minh Le Duc** (Lê Đức Minh). Each post that runs an experiment gets a folder here, so a reader can rerun the numbers instead of trusting them.
23
+
24
+ The blog covers LLM agents, retrieval and memory, Vietnamese NLP, encoder benchmarking and small-model evaluation. New posts go to ZyVOP first; older ones are on Medium.
25
+
26
+ ## Contents
27
+
28
+ | Folder | File | Used by |
29
+ | :--- | :--- | :--- |
30
+ | `agent-memory-tax/` | `longmemeval_oracle.json` | Draft post "Three Papers Say Your Agent's Memory Loses to Plain RAG" (a laptop test of where agent memory loses to plain retrieval). The link will be added here once it is live. |
31
+
32
+ More folders are added as posts ship.
33
+
34
+ ## `agent-memory-tax/longmemeval_oracle.json`
35
+
36
+ The **oracle** split of LongMemEval (cleaned release): 500 questions, each paired only with the chat sessions that contain the evidence. It is the "easy retrieval" setting: no distractor sessions, so it isolates what a memory layer does to the evidence rather than whether it can find it.
37
+
38
+ **Provenance.** This is a byte-for-byte copy of `longmemeval_oracle.json` from [`xiaowu0162/longmemeval-cleaned`](https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned) (SHA-256 `821a2034d219ab45846873dd14c14f12cfe7776e73527a483f9dac095d38620c`, 15,388,478 bytes). Nothing was filtered, rewritten or re-labelled. It is mirrored here only so the blog's scripts point at one stable place.
39
+
40
+ **Fields** (one JSON object per question):
41
+
42
+ | Field | Meaning |
43
+ | :--- | :--- |
44
+ | `question_id` | Question identifier |
45
+ | `question_type` | One of six types (counts below) |
46
+ | `question` / `answer` | The question and its gold answer |
47
+ | `question_date` | When the question is asked in the simulated timeline |
48
+ | `haystack_sessions` | The chat sessions (lists of `role` / `content` turns) |
49
+ | `haystack_session_ids` / `haystack_dates` | Session ids and timestamps, aligned with `haystack_sessions` |
50
+ | `answer_session_ids` | Which sessions hold the evidence |
51
+
52
+ **Question types:** `temporal-reasoning` 133, `multi-session` 133, `knowledge-update` 78, `single-session-user` 70, `single-session-assistant` 56, `single-session-preference` 30.
53
+
54
+ ```python
55
+ import json
56
+ from huggingface_hub import hf_hub_download
57
+
58
+ path = hf_hub_download("LakoreAI/experiments", "agent-memory-tax/longmemeval_oracle.json", repo_type="dataset")
59
+ data = json.load(open(path))
60
+ print(len(data), data[0]["question_type"])
61
+ ```
62
+
63
+ The dataset viewer is off because `haystack_sessions` is deeply nested; load the JSON directly.
64
+
65
+ ## License and attribution
66
+
67
+ MIT, inherited from the upstream release. All credit for the benchmark goes to its authors. If you use the data, cite the original paper, not this mirror:
68
+
69
+ > Wu et al., *LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory*, 2024. https://arxiv.org/abs/2410.10813
70
+
71
+ ## Where the blog lives
72
+
73
+ - **ZyVOP** (canonical for new posts): https://zyvop.com/author/l445
74
+ - **Medium** (legacy): https://medium.com/@minhle_0210
75
+ - **GitHub:** https://github.com/MinLee0210
76
+ - **LinkedIn:** https://www.linkedin.com/in/minhle007/
77
+
78
+ ### ZyVOP (13 posts)
79
+
80
+ | Post | Date |
81
+ | :--- | :--- |
82
+ | [BM25 Beat the RAG Frameworks on Vietnamese Law](https://zyvop.com/bm25-beat-the-rag-frameworks-on-vietnamese-law-iihod) | 2026-10-23 |
83
+ | [Your Agent Has 20 Skills. Does It Load the Right One?](https://zyvop.com/your-agent-has-20-skills-does-it-load-the-right-one-qp5u2) | 2026-10-22 |
84
+ | [A Model Asked Itself Permission, Then Spent Test ETH](https://zyvop.com/a-model-asked-itself-permission-then-spent-test-eth-5rw0h) | 2026-10-21 |
85
+ | [Jev-as-Guardrail (JAG): What Typed Questions Actually Buy You](https://zyvop.com/jev-as-guardrail-jag-what-typed-questions-actually-buy-you-padzh) | 2026-10-18 |
86
+ | [Fun Project: I spent a weekend teaching my browser how wind works](https://zyvop.com/fun-project-i-spent-a-weekend-teaching-my-browser-how-wind-works-3y16r) | 2026-10-11 |
87
+ | [Jev's Boring Use Cases Are the Ones That Actually Work](https://zyvop.com/jev-s-boring-use-cases-are-the-ones-that-actually-work-z8lqo) | 2026-09-27 |
88
+ | [I Reproduced RLCD to Find Out What Its "Calibrated RL" Actually Does](https://zyvop.com/i-reproduced-rlcd-to-find-out-what-its-calibrated-rl-actually-does-jrv0k) | 2026-09-26 |
89
+ | [Inside RLCD: The Estimator, the Proof, and the 30 Runs That Changed My Mind](https://zyvop.com/inside-rlcd-the-estimator-the-proof-and-the-30-runs-that-changed-my-mind-0qzv9) | n/d |
90
+ | [I Took Laya's Calibrated RL Apart: It's Cross-Entropy With Extra Steps](https://zyvop.com/i-took-laya-s-calibrated-rl-apart-it-s-cross-entropy-with-extra-steps-vtl6r) | n/d |
91
+ | [Understanding RLCD: Calibration Techniques for Decision-Only LLMs](https://zyvop.com/understanding-rlcd-calibration-techniques-for-decision-only-llms-bruse) | n/d |
92
+ | [Jev, Explained: The AI That Refuses to Write a Word](https://zyvop.com/jev-explained-the-ai-that-refuses-to-write-a-word-w98gs) | n/d |
93
+ | [Pi-warden: Using Jev to Block Destructive Commands in 48 Hours](https://zyvop.com/pi-warden-using-jev-to-block-destructive-commands-in-48-hours-5lfoy) | n/d |
94
+ | [How to Build a Sustainable Technical Writing Habit](https://zyvop.com/how-to-build-a-sustainable-technical-writing-habit-pp8oz) | n/d |
95
+
96
+ ### Medium (49 posts)
97
+
98
+ Published on `medium.com/@minhle_0210` and the *Artificial Intelligence in Plain English* publication. Dates marked `n/d` were not recorded. Taken from the author's own index, which is a floor rather than a complete list.
99
+
100
+ | Post | Date |
101
+ | :--- | :--- |
102
+ | [Prompt, Context, Harness, Loop: An Agent's Anatomy](https://medium.com/@minhle_0210/prompt-context-harness-loop-an-agents-anatomy-642db41429fb) | 2026-07-05 |
103
+ | [AI Agent Guardrails: Why Ontologies Beat Prompting](https://medium.com/@minhle_0210/ai-agent-guardrails-why-ontologies-beat-prompting-d460942ddf92) | 2026-08-04 |
104
+ | [Why Your RAG Can't Answer "Why" — Knowledge Graphs, Pt 1](https://medium.com/@minhle_0210/why-your-rag-cant-answer-why-knowledge-graphs-pt-1-a92d1fc4f2d1) | 2026-08-06 |
105
+ | [Building a GraphRAG Pipeline — What It Really Costs, Pt 2](https://ai.plainenglish.io/building-a-graphrag-pipeline-what-it-really-costs-pt-2-13e7e3541d9f) | 2026-08-12 |
106
+ | [Context Graphs: Agent Memory That Remembers Why — Pt 3](https://ai.plainenglish.io/context-graphs-agent-memory-that-remembers-why-pt-3-29334c9c1241) | 2026-08-13 |
107
+ | [From RAG to CAG: A Smarter Approach to AI](https://ai.plainenglish.io/from-rag-to-cag-a-smarter-approach-to-ai-c0a9f21cd6a0) | n/d |
108
+ | [5 ways to improve your RAG — Part 1: Chunking strategies](https://ai.plainenglish.io/5-ways-to-improve-your-rag-part-1-chunking-strategies) | n/d |
109
+ | [The Architecture Behind a Closed-Loop Eval System for a Multimodal Agent](https://ai.plainenglish.io/the-architecture-behind-a-closed-loop-eval-system-for-a-multimodal-agent-edacf586c8b4) | 2026-07-26 |
110
+ | [Breaking down Speculative RAG](https://medium.com/@minhle_0210/breaking-down-speculative-rag-933365f289fa) | n/d |
111
+ | [Do some search tools for LLM](https://ai.plainenglish.io/do-some-search-tools-for-llm-0deffbd27470) | 2025-05-13 |
112
+ | [Open-Source Intelligence](https://medium.com/@minhle_0210/open-source-intelligence-5f65fb6c2257) | n/d |
113
+ | [What DeepSeek's Open-Source Agent Harness Gets Right](https://medium.com/@minhle_0210/what-deepseeks-open-source-agent-harness-gets-right-b85f57533802) | 2026-08-19 |
114
+ | [Agent harness or agent framework? They're not the same](https://ai.plainenglish.io/agent-harness-or-agent-framework-theyre-not-the-same-3d41caa5aeaa) | 2026-09-05 |
115
+ | [Deterministic Agent Runtime with Context Pruning](https://ai.plainenglish.io/deterministic-agent-runtime-with-context-pruning-448be2ed2173) | 2026-09-16 |
116
+ | [Repo-to-Skill Framework Boosts Agent Performance](https://medium.com/@minhle_0210/repo-to-skill-framework-boosts-agent-performance-27430534121a) | 2026-09-22 |
117
+ | [Jev, Explained: The AI That Refuses to Write a Word](https://medium.com/@minhle_0210/jev-explained-the-ai-that-refuses-to-write-a-word-b1788b932ae0) | 2026-09-24 |
118
+ | [5 Tiny Language Models for Tool Calling (Part 3)](https://medium.com/@minhle_0210/5-tiny-language-models-for-tool-calling-part-3-ebcda32c2518) | 2026-08-18 |
119
+ | [How Kimi K3 reads a million tokens](https://medium.com/@minhle_0210/how-kimi-k3-reads-a-million-tokens-f2dc7c70ceff) | 2026-08-16 |
120
+ | [5 Tiny Language Models You Need to Know in 2026 (Part 2)](https://ai.plainenglish.io/5-tiny-language-models-you-need-to-know-in-2026-part-2-cafb5b04a022) | 2026-08-10 |
121
+ | [Kimi K3 has 2.8T parameters. Moving them was the hard part.](https://medium.com/@minhle_0210/kimi-k3-has-2-8t-parameters-moving-them-was-the-hard-part-617e0ee7a97a) | 2026-08-09 |
122
+ | [I Spent My Weekend Benchmarking Vietnamese BERT Models (So You Don't Have To)](https://ai.plainenglish.io/i-spent-my-weekend-benchmarking-vietnamese-bert-models-so-you-dont-have-to-0d19aa280736) | 2026-07-04 |
123
+ | [Weekend NLP: Fine-tuning Transformers for Language Detection with HuggingFace](https://ai.plainenglish.io/weekend-nlp-fine-tuning-transformers-for-language-detection-with-huggingface-4041ac9ca20e) | n/d |
124
+ | [5 Tiny Language models that you need to know (Part 1)](https://ai.plainenglish.io/5-tiny-language-models-that-you-need-to-know-part-1-f93fb05cbffa) | n/d |
125
+ | [On short of "TinyLlama: An Open-Source Small Language Model"](https://ai.plainenglish.io/on-short-of-tinyllama-an-open-source-small-language-model-66e747f181fb) | 2024-03-12 |
126
+ | [KeyBERT: Effortless Keyword Extraction](https://ai.plainenglish.io/keybert-effortless-keyword-extraction-c0ac001ceb19) | 2024-07 (~) |
127
+ | [Introduction to series "🕹 Fun Project:…"](https://medium.com/@minhle_0210/introduction-to-series-fun-project-d01905d05804) | n/d |
128
+ | [Fun Project: Video Chat with SmolVLM](https://ai.plainenglish.io/fun-project-video-chat-with-smolvlm-32ef19e6cc42) | 2025-06-16 |
129
+ | [Fun Project: IELTS Suggestor](https://ai.plainenglish.io/fun-project-ietls-suggestor-3c0b35d24218) | n/d |
130
+ | [🕹 Fun Project: Treat Llama-3 as Coding Agent 🤖](https://medium.com/@minhle_0210/fun-project-treat-llama-3-as-coding-agent-bb4e715c2aa7) | n/d |
131
+ | [Fun Project: Data Exploration on a Saturday Morning](https://medium.com/@minhle_0210/fun-project-data-exploration-on-a-sunday-morning-560fb0ae6e1b) | 2026-04 |
132
+ | [Fun Project: I Built a Compressor That Thinks in Tokens](https://medium.com/@minhle_0210/fun-project-i-built-a-compressor-that-thinks-in-tokens-cd16c1d0a503) | 2026-09-05 |
133
+ | [Fun Project: Evaluating Vietnamese Embedding Models](https://ai.plainenglish.io/fun-project-evaluating-vietnamese-embedding-models) | n/d |
134
+ | [Top-K at Scale: A Real-Time Ranking System Walkthrough](https://ai.plainenglish.io/top-k-at-scale-a-real-time-ranking-system-walkthrough-5708fd20f6f2) | 2026-08-05 |
135
+ | [How do you deduplicate a stream you don't control?](https://medium.com/@minhle_0210/how-do-you-deduplicate-a-stream-you-dont-control-6862bbaaa3d9) | 2026-08-20 |
136
+ | [Why committing Kafka offsets out of order loses data](https://medium.com/@minhle_0210/why-committing-kafka-offsets-out-of-order-loses-data-ef9b8eb0c603) | 2026-08-21 |
137
+ | [The backlog that killed my Kafka consumer](https://medium.com/@minhle_0210/the-backlog-that-killed-my-kafka-consumer-2d6f22b9eff6) | 2026-08-22 |
138
+ | [Why your coding agent's bill grows faster than the chat](https://medium.com/@minhle_0210/why-your-coding-agents-bill-grows-faster-than-the-chat-2d9ecb37423d) | 2026-09-01 |
139
+ | [Five ways to invalidate your prompt cache](https://medium.com/@minhle_0210/five-ways-to-invalidate-your-prompt-cache-94fd845ec546) | 2026-09-02 |
140
+ | [Dynamic Reasoning Budgets Cut Inference Costs](https://ai.plainenglish.io/dynamic-reasoning-budgets-cut-inference-costs-51c36c442227) | 2026-09-17 |
141
+ | [What Video Summarization Actually Looks Like (And Why It's So Hard)](https://medium.com/@minhle_0210/what-video-summarization-actually-looks-like-and-why-its-so-hard-a7a6a92b18e9) | 2026-07-26 |
142
+ | [Computer Vision with UAV, Applications and Futures](https://medium.com/@minhle_0210/computer-vision-with-uav-applications-and-futures-ca197901dd01) | 2026-06-01 |
143
+ | [SmolVLM2 vs Qwen2.5-VL: Real-Time RTSP Edge Video Summarization Under 8GB VRAM](https://medium.com/@minhle_0210/smolvlm2-vs-qwen2-5-vl-real-time-rtsp-edge-video-summarization-under-8gb-vram-d10df953d803) | 2026-09-23 |
144
+ | [Kalman Filter: From Concepts to Applications](https://ai.plainenglish.io/kalman-filter-from-concepts-to-applications-f35484e54822) | 2025-05-01 |
145
+ | [What do I do as an AI Engineer](https://ai.plainenglish.io/what-do-i-do-as-an-ai-engineer-cfb79fd818a1) | n/d |
146
+ | [The Hard Parts of Being an AI Engineer](https://medium.com/@minhle_0210/the-hard-parts-of-being-an-ai-engineer-4fe72dcfcfd8) | n/d |
147
+ | [Picking the Right Programming Language Actually Matters](https://medium.com/@minhle_0210/picking-the-right-programming-language-actually-matters-52678e12433f) | n/d |
148
+ | [Intro to my new series "Back to Basics: …"](https://medium.com/@minhle_0210/intro-to-my-new-series-back-to-basics-979040b158ca) | n/d |
149
+ | [10 Dunder Methods Every Python Developer Should Know! 🚀](https://medium.com/@minhle_0210/10-dunder-methods-every-python-developer-should-know-fe2a8c6d8445) | n/d |
150
+ | [🕯️ Candle thought: Can AI replace human?](https://medium.com/@minhle_0210/%EF%B8%8F-candle-light-can-ai-replace-human-2f2bf15c1387) | n/d |