← back to report

rasbt/LLMs-from-scratch

// all events — 2026-05-25

Type AI Score Description Actor Reason Date
commit 0.40 Add some appendix E runtimes (#927) Repetitive commit lines suggest possible 2025-12-19
commit 0.30 Add Tiny Aya from scratch (#962) Standard PR title format, but too generi 2026-02-19
commit 0.30 Readability and code quality improvements (#959) Generic but plausible human commit messa 2026-02-18
commit 0.30 Sliding window KV Cache bug fix (#925) Structured but human bug report; slight 2025-12-16
commit 0.30 Add Olmo 3 README (#915) README update with repetitive edits; min 2025-11-23
commit 0.20 User argpars utils to show default args on command line Brief technical command line utility des 2026-03-02
commit 0.20 remove redundant assignment (#961) Brief technical note, typical human comm 2026-02-19
commit 0.20 Use correct input in layernorm example (#960) Bullet points follow human commit habit, 2026-02-19
commit 0.20 Fix flex attention in PyTorch 2.10 (#957) Specific technical fix referencing PyTor 2026-02-09
commit 0.20 Fix docstring parameter names in compute_dpo_loss function ( Technical docstring fix, jargon present 2026-01-29
commit 0.20 Update unit tests for CI (#952) CI update commits often terse; bullet fo 2026-01-27
commit 0.20 Cover Python 3.12 (#933) Simple PR title referencing issue; typic 2025-12-27
commit 0.20 Gated DeltaNet updates (#926) Concise, technical PR title; human-autho 2025-12-19
commit 0.20 n_heads × d_head -> d_head × d_head in DeltaNet (#903) Structured explanation but uses human-li 2025-11-06
commit 0.20 Training on MPS in PyTorch 2.9 (#900) Terse, informal update; typical human co 2025-11-01
commit 0.20 simplify uv command (#898) Very brief and casual; typical human com 2025-11-01
commit 0.20 Add bonus dependencies to pyproject (#897) Minimal, informal updates; typical human 2025-10-29
commit 0.20 Fix ffn link (#892) Simple fix with manual attribution; natu 2025-10-22
commit 0.20 Make quote style consistent (#891) Concise, direct change description; typi 2025-10-22
commit 0.20 - docs(moe): correct arXiv link for DeepSeekMoE (#890) Structured, factual documentation correc 2025-10-21
commit 0.20 Mixture-of-Experts intro (#888) Brief, descriptive title; typical human 2025-10-20
commit 0.20 Make it easier to toggle between thinking and instruct varia Clear, direct feature improvement; natur 2025-10-17
commit 0.20 Update the compression rate comment in MLA (#883) Informal comment update; typical human r 2025-10-14
pr 0.20 Add cross-layer KV sharing Brief, technical description with minor 2026-05-17
commit 0.10 Tips on forking and updating this repo locally (#1027) Informal, concise commit messages with t 2026-05-17
commit 0.10 Add cross-layer KV sharing (#1025) Commit messages are terse and domain-spe 2026-05-17
commit 0.10 fix: pin 1 unpinned action(s) (#987) Template-based automated security fix no 2026-03-26
commit 0.10 fix: added KVcache in `generate_text_basic_stream` (#981) Concise technical commit message with sp 2026-03-21
commit 0.10 Minor typo fix (#974) Simple typo fix notice, typical human co 2026-03-07
commit 0.10 Bpe whitespace fixes (#975) Brief technical reference to BPE whitesp 2026-03-07
commit 0.10 Add more analysis to qwen3.5 image Concise, terse enhancement description t 2026-03-04
commit 0.10 Use full HF url Very brief technical instruction with ab 2026-03-03
commit 0.10 Qwen3.5 from scratch (#969) Terse commit title with minimal, human-l 2026-03-03
commit 0.10 Jupyter scrolling glitch tips (#965) Specific issue reference to Jupyter scro 2026-02-27
commit 0.10 image size Extremely terse; typical human GitHub sh 2026-02-19
commit 0.10 image size Very minimal, template-like entry 2026-02-19
commit 0.10 formatting fix Too brief to assess; likely human short 2026-02-19
commit 0.10 yearly update Brief, minimal human phrase with no AI h 2026-01-02
commit 0.10 upload saved nb Terse, informal human note; likely manua 2025-12-21
commit 0.10 Update memory efficient loading nb Direct technical note; common human shor 2025-12-21
commit 0.10 update submodule Minimal, human-specific action; no AI ph 2025-12-20
commit 0.10 Remove persistent flag from cache buffers (#916) Direct technical title; typical human ch 2025-11-25
commit 0.10 Update README wrt multi-query attention Formal but clear technical summary, not 2025-11-17
commit 0.10 Fix MHAEinsum weight dimension bug when d_in != d_out (#857) Detailed technical fix with unit tests; 2025-11-01
commit 0.10 Windows compile (#845) Minimal, terse commit-like entries show 2025-09-26
commit 0.10 Update package dependencies (#842) Single-line PR title is typical GitHub s 2025-09-22
commit 0.10 Improve MoE implementation (#841) Brief title only; no text shows AI styli 2025-09-22
commit 0.10 Note about devcontainer root usage (#833) Very minimal entry with template placeho 2025-09-21
commit 0.10 Note about RoPE usage (#839) Terse bullet points with simple notes; n 2025-09-20
pr 0.10 Add Deepseek Sparse Attention (DSA) implementation (ch04/09_ PR description is technical, precise, an 2026-04-17
pr 0.10 Add troubleshooting guide for common questions Very brief and generic; no AI stylistic 2026-05-11
pr 0.10 fix: remove unsafe exec() in ch06.py Template structure, but free-text is tec 2026-05-04
pr 0.10 fix: handle token_ids length<2 in find_freq_pair Concise, domain-specific, and typical hu 2026-04-11
pr 0.10 Handle short-token edge case in BPE pair selection Structured, but with concise humanlike c 2026-04-09
pr 0.10 Clarify GELU input example by using zero-centered random inp Technical, domain-specific language with 2026-04-08
pr 0.10 uv packages must be 7 days old or older Brief, domain-specific changelog, no AI 2026-03-31
pr 0.10 Swap urllib.request with requests Casual phrasing with minor mistakes, hum 2026-03-31
pr 0.08 Muon optimizer Brief, informal free-text with project-s 2026-05-19
commit 0.05 Muon optimizer (#1029) Commit messages are terse and use domain 2026-05-19
commit 0.00 Add Deepseek Sparse Attention (DSA) implementation (ch04/09_ Commit messages are terse and contain do 2026-05-23
commit 0.00 Fix TensorFlow dependency marker for Linux ARM (#1028) No free-text content; only title with is 2026-05-19
commit 0.00 Unique names to improve automerge if all tests pass (#1026) No free-text content; only template/PR t 2026-05-17
commit 0.00 Add troubleshooting guide for common questions (#1024) Standard commit message with no AI hallm 2026-05-11
commit 0.00 Fix typo in README for KV-cache totals (#1023) Brief, typo fix message with domain abbr 2026-05-06
commit 0.00 Fix SWA (#1013) Very terse commit message; no AI signals 2026-04-16
commit 0.00 Fix BPE empty-pair edge case in find_freq_pair (#1004) Standard commit message with co-author, 2026-04-11
commit 0.00 Gemma 4 (#1005) Very brief PR title, no signs of AI gene 2026-04-11
commit 0.00 Some gemma 3 improvements (#1000) Terse commit message; informal tone and 2026-04-06
commit 0.00 Remove 7 days requirement to improve windows compatibility ( Direct, terse commit message with no AI 2026-04-04
commit 0.00 uv packages must be 7 days old (#994) Terse, informal commit message; no AI ha 2026-04-01
commit 0.00 Swap urllib.request with requests (#993) Concise commit title, typical human styl 2026-03-31
commit 0.00 harded the link checker Terse human message with a typo ('harded 2026-03-07
commit 0.00 chore: Update outdated GitHub Actions versions (#951) Standard chore commit with minimal human 2026-01-19
commit 0.00 link GRPO notebook (#950) Brief, terse human message typical for G 2026-01-18
commit 0.00 Optional weight tying for Qwen3 and Llama3.2 pretraining (#9 Technical domain jargon and casual 'typo 2026-01-14
commit 0.00 Fix encoding of multiple preceding spaces in BPE tokenizer. Structured human PR with specific techni 2026-01-10
commit 0.00 Chapter 5 with alternative LLMs (Qwen3, Llama 3) (#943) Human PR with iterative technical fixes 2026-01-09
commit 0.00 Correct batch_idx in appendix A logging (#942) Technical human fix with specific batch_ 2026-01-08
commit 0.00 Fix Olmo3 YaRN RoPE implementation bug (#940) Specific technical bug fix with model na 2026-01-04
commit 0.00 Correct 'pix' to 'pixi' in README.md (#935) Brief human typo correction commit with 2026-01-02
commit 0.00 Clean up native-uv.md documentation (#938) Minimal human-style documentation cleanu 2026-01-02
commit 0.00 Fix GitHub CI timeout issue for link checker (#937) Specific CI troubleshooting with iterati 2026-01-02
commit 0.00 Olmo 3 from scratch (#914) Contains generic commit titles with mini 2025-11-23
commit 0.00 RoPE decay plot (#910) Simple, repetitive commit titles, showin 2025-11-17
commit 0.00 Write-up on how to get the most out of this book (#909) Minimal description, typical of a simple 2025-11-13
commit 0.00 fix(GatedDeltaNet): Init param A from log of a uniform distr Concise, direct commit-message style wit 2025-11-09
commit 0.00 Use consistent title case Very brief, action-oriented phrase, typi 2025-11-06
commit 0.00 Fix empty device issue (#904) Direct fix description, common in human 2025-11-06
commit 0.00 Image resizing Extremely brief, generic title, likely h 2025-11-03
commit 0.00 Gated DeltaNet write-up (#901) Mix of simple titles and mundane commit 2025-11-03
commit 0.00 Use figure numbers in ch05-7 (#881) Terse technical commit messages with dom 2025-10-13
commit 0.00 Add alternative attention structure (#880) Brief, informal title typical of GitHub 2025-10-13
commit 0.00 sliding window attention (#879) Minimal technical title without AI hallm 2025-10-13
commit 0.00 Add other appendices for completeness (#878) Template-structured PR with minimal, ter 2025-10-13
commit 0.00 rm plot Extremely terse command-like commit mess 2025-10-12
commit 0.00 Multi-Head Latent Attention (#876) Template-structured PR with brief techni 2025-10-12
commit 0.00 Use GB instead of GiB consistently (#875) Concise technical specification change. 2025-10-11
commit 0.00 Grouped-Query Attention memory (#874) Template PR with action-oriented, concis 2025-10-11
commit 0.00 Use inference_device Very brief, technical command-like phras 2025-10-09
commit 0.00 Add simpler BPE, and make previous BPE better (#870) Template-structured PR with brief techni 2025-10-09
commit 0.00 Qwen3 and evaluation bonus materials (#869) Title and branch naming show informal, h 2025-10-08
commit 0.00 Switch from urllib to requests to improve reliability (#867) Brief, terse commit messages with minima 2025-10-07
commit 0.00 Add missing comma in imports in README (#865) Short, direct title for a minor document 2025-10-06
commit 0.00 Note about output dimensions (#862) Succinct title about documentation lacks 2025-10-01
commit 0.00 Update ollama address (#861) Title is a simple, straightforward updat 2025-10-01
commit 0.00 some typo fixes (#858) Informal 'some typo fixes' and specific 2025-09-30
commit 0.00 Test dependencies with Python 3.13 (#843) Title is direct technical work; commit m 2025-09-27
commit 0.00 Update generate script (#847) Title and iterative update commits show 2025-09-27
commit 0.00 Numerically stable generate on mps (#849) Technical title and brief, specific comm 2025-09-27
commit 0.00 Requirements update (#851) Title and commit about 'tricker workers' 2025-09-27
pr 0.00 Fix TensorFlow dependency marker for Linux ARM No free-text content; only title, human 2026-05-19
pr 0.00 Tips on forking and updating this repo locally Only template fields and issue/link, no 2026-05-17
pr 0.00 Unique names to improve automerge if all tests pass Only a title; no free-text content to ev 2026-05-17
pr 0.00 Fix typo in README for KV-cache totals Concise, domain-specific phrasing lackin 2026-05-04
pr 0.00 Fix SWA Minimal, issue-referencing title; no AI 2026-04-16
pr 0.00 Add Gemma 4 E2B and E4B (from scratch) Standard PR title, no AI indicator or at 2026-04-11
pr 0.00 Some gemma 3 improvements Succinct and informal, not showing AI ge 2026-04-06
pr 0.00 Remove 7 days requirement to improve windows compatibility Standard issue fix message; concise, no 2026-04-04
pr 0.00 Debug exclude-newer Casual explanation, domain-specific cont 2026-04-04
pr 0.00 Fix the ssl issue for MacOS M1 when downloading gpt model re Concise, specific issue summary and erro 2026-03-29
pr 0.00 Actions Single keyword title lacks AI hallmarks; 2026-03-30
pr 0.00 fix: pin 1 unpinned action(s) — 2026-03-26
pr 0.00 fix: `qwen3.5-plus-kv-cache.ipynb` to use KVCache — 2026-03-17
pr 0.00 [Invalid] Mistake — 2026-03-08
pr 0.00 Minor typo fix — 2026-03-07
pr 0.00 Bpe whitespace fixes — 2026-03-07
pr 0.00 Qwen3.5 from scratch — 2026-03-03
pr 0.00 Jupyter scrolling glitch tips — 2026-02-27
pr 0.00 Add Tiny Aya from scratch — 2026-02-19
pr 0.00 Remove redundant model assignment — 2026-02-19
pr 0.00 Use correct input in layernorm example — 2026-02-19
pr 0.00 Readability and code quality improvements — 2026-02-18
pr 0.00 Feature/tool calling experiment — 2026-02-15
pr 0.00 Fix flex attention in PyTorch 2.10 — 2026-02-09
pr 0.00 Download — 2026-02-01
pr 0.00 Fix docstring parameter names in compute_dpo_loss function — 2026-01-29
pr 0.00 Update unit tests for CI — 2026-01-27
pr 0.00 chore: Update outdated GitHub Actions versions — 2026-01-19
pr 0.00 Link GRPO notebook — 2026-01-18
pr 0.00 Optional weight tying for Qwen3 and Llama3.2 pretraining — 2026-01-14
pr 0.00 Suggestion: Add quick start example for training and inferen — 2026-01-11
pr 0.00 Fix encoding of multiple preceding spaces in BPE tokenizer. — 2026-01-10
pr 0.00 Chapter 5 with alternative LLMs (Qwen3, Llama 3) — 2026-01-09
pr 0.00 Correct batch_idx in appendix A logging — 2026-01-05
pr 0.00 Fix Olmo3 YaRN RoPE implementation bug — 2026-01-03
pr 0.00 Correct 'pix' to 'pixi' in README.md — 2026-01-02
pr 0.00 Clean up native-uv.md documentation — 2026-01-02
pr 0.00 Fix GitHub CI timeout issue for link checker — 2026-01-02
pr 0.00 Cover Python 3.12 — 2025-12-27
pr 0.00 Add some appendix E runtimes — 2025-12-19
pr 0.00 Added comments in class SimpleTokenizerV1 — 2025-11-27
pr 0.00 Gated DeltaNet updates — 2025-12-17
pr 0.00 Optimized KV Cache (sliding window) bug fix — 2025-12-15
pr 0.00 Sidd — 2025-12-03
pr 0.00 GatedDeltaNet code: Initialize A as the log of a uniform dis — 2025-11-08
pr 0.00 Remove persistent flag from cache buffers — 2025-11-24
pr 0.00 Add Olmo 3 README — 2025-11-23
pr 0.00 Olmo 3 from scratch — 2025-11-23
pr 0.00 Fix empty device issue — 2025-11-06
pr 0.00 RoPE decay plot — 2025-11-17
pr 0.00 Write-up on how to get the most out of this book — 2025-11-13
pr 0.00 n_heads × d_head -> d_head × d_head in DeltaNet — 2025-11-06
pr 0.00 Gated DeltaNet write-up — 2025-11-03
pr 0.00 Training on MPS in PyTorch 2.9 — 2025-11-01
pr 0.00 Add bonus dependencies to pyproject — 2025-10-29
pr 0.00 Fix MHAEinsum weight dimension bug when d_in != d_out (#857) — 2025-10-22
pr 0.00 Simplify uv command — 2025-11-01
pr 0.00 some typo fixes — 2025-09-30
pr 0.00 docs(MoE): minor links fix — 2025-10-20
pr 0.00 Fix ffn link — 2025-10-22
pr 0.00 Style consistency update — 2025-10-21
pr 0.00 Mixture-of-Experts intro — 2025-10-20
pr 0.00 Make it easier to toggle between thinking and instruct varia — 2025-10-17
pr 0.00 Update the compression rate comment in MLA — 2025-10-14
pr 0.00 Use figure numbers in ch05-7 — 2025-10-13
pr 0.00 Add alternative attention structure — 2025-10-13
pr 0.00 Sliding window attention — 2025-10-13
pr 0.00 Add other appendices for completeness — 2025-10-12
pr 0.00 Multi-Head Latent Attention — 2025-10-12
pr 0.00 Use GB instead of GiB consistently — 2025-10-11
pr 0.00 Grouped-Query Attention memory — 2025-10-11
pr 0.00 Add simpler BPE, and make previous BPE better — 2025-10-09
pr 0.00 Qwen3 and evaluation bonus materials — 2025-10-08
pr 0.00 Switch from urllib to requests to improve reliability — 2025-10-07
pr 0.00 Update Docker file — 2025-10-06
pr 0.00 Add missing comma in imports in README — 2025-10-06
pr 0.00 docs: Auto-translate README and Wiki — 2025-10-05
pr 0.00 Fix consistency in the definition of `MultiHeadAttentionWrap — 2025-09-30
pr 0.00 Note about output dimensions d_out in ch03 — 2025-10-01
pr 0.00 Update ollama address — 2025-10-01
pr 0.00 Test dependencies with Python 3.13 — 2025-09-23
pr 0.00 Update generate script — 2025-09-27
pr 0.00 Numerically stable generate on mps — 2025-09-27
pr 0.00 Requirements update — 2025-09-27
pr 0.00 Requirements update — 2025-09-27
pr 0.00 Windows compile — 2025-09-26
pr 0.00 `Qwen3Tokenizer` fix for Qwen3 Base models and generation mi — 2025-09-15
pr 0.00 Update package dependencies — 2025-09-22
pr 0.00 Improve MoE implementation — 2025-09-22
pr 0.00 Note about devcontainer root usage — 2025-09-16