11月12月1月2月3月4月5月6月7月8月9月10月
周一周三周五
一年内 256781 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/nccl-only 分支比较 10 提交 »f19202c858 small fixes on top of @chinthysl PR to kick off the feature branch10e4dac2db Merge branch 'multinode-slurm' of https://github.com/ahrefs/llm.c into ahrefs-multinode-slurm2ddf128365 Merge pull request #615 from gordicaleksa/relax_grad_tests8079bd9dbb Relax grad tensor thresholds a bitfd564f7b8a add the gpt2 1558M repro script too
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 2.5.x 分支比较 3 提交 »cd67ccc6ef Revert "[submodule] update libxsmm to latest" (#7854)25a680d294 [submodule] update libxsmm to latest (#7853)30cfa6ea54 [release] bump version to 2.5.0
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/managed2 分支比较 10 提交 »18298f3a40 i misspelled reduced8c586f9158 reduce across GPUs nicere6856bc566 fallback to memory allocation of m,v,master_weights on host automatically in case of OOM. will run slower but won't OOM2882ec6b9e fix makefile for multigpu setupsf72c1f2c64 hint to host
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 2.3.x 分支比较 2 提交 »35264b07b9 [release] bump version to 2.3.1 (#7533)4aa02d855c [release] update version (#7488)
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 2.4.x 分支c6c874bf7e [2.4.0] Release branch cut. (#7738)
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 2.1.x 分支比较 6 提交 »1547bd931d [doc] use tqdm from tqdm.auto (#7191)69247f5b03 [doc] do not display nn.tensorflow and nn.mxnet (#7190)64a8a0e3cc [GraphBolt] Slight `gb.expand_indptr` improvement on CPU. (#7189)9e8a308e52 [doc] install cpu version by default in notebooks (#7186)2550ac5c91 [GraphBolt][PyG] Modify PyG example with `to_pyg_data` (#7123)
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/lr_schedulers 分支比较 10 提交 »dbe06082ca delete spurious check its fine4b5fecc8b2 small touchups and add the wsd schedule96ebb16298 simplify lr schedulers9abd9ec8a8 Merge branch 'add_lr_schedules' of https://github.com/gordicaleksa/llm.c into gordicaleksa-add_lr_schedulesdad58b6800 Merge pull request #625 from gordicaleksa/add_nccl_to_readme
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/libdir 分支比较 10 提交 »22d1607f6d delete these files, they are now in scripts/ folder and betterfc40ffc4aa adjust readme to new scripts/ dira0d5dbbda6 scripts folder35aa110244 more refactorf8f9e49ba3 Merge branch 'master' into feature/libdir
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/load_bf16 分支比较 10 提交 »9d70d9a1b2 profile and test only use bf16. but the train script can be run with fp32 or bf16 or fp16. fp16 will error, thougha3f5ad9095 reshuffle the ifdefs to make bf16 the default if no PRECISION is requested via defines4f7d8d9f44 allow user to make different precisions, add prints and error handling around precisions82d7907214 adjust comment9a91b40207 bring back original ordering. i also had to bump the thresholds by 3X for some tensors and i don't exactly know why sad
