tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Built site for gh-pages

Browse Source
This commit is contained in:
Quarto GHA Workflow Runner
2025-08-08 11:51:10 +00:00
parent 26f906cd89
commit ceb67f0b89
4 changed files with 213 additions and 200 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
search.json
View File

@@ -1632,7 +1632,7 @@
"href": "docs/nd_parallelism.html#examples",
"title": "N-D Parallelism (Beta)",
"section": "Examples",
"text": "Examples\n\nHSDP on 2 nodes with 4 GPUs each (8 GPUs total):\n\nYou want FSDP within each node and DDP across nodes.\nSet dp_shard_size: 4 and dp_replicate_size: 2.\n\nFSDP + TP on a single 8-GPU node:\n\nYou want to split the model across 4 GPUs using FSDP, and further split each layer across 2 GPUs with TP.\nSet dp_shard_size: 4 and tensor_parallel_size: 2.\n\nFSDP + CP on a single 8-GPU node for long context:\n\nYou want to shard the model across all 8 GPUs and also split the sequence length across all 8 GPUs.\nSet dp_shard_size: 8 and context_parallel_size: 8. Note: this means the data parallel group and context parallel group are the same. A more common setup might be to shard across a smaller group.",
"text": "Examples\n\n\n\n\n\n\nTip\n\n\n\nSee our example configs here.\n\n\n\nHSDP on 2 nodes with 4 GPUs each (8 GPUs total):\n\nYou want FSDP within each node and DDP across nodes.\nSet dp_shard_size: 4 and dp_replicate_size: 2.\n\nFSDP + TP on a single 8-GPU node:\n\nYou want to split the model across 4 GPUs using FSDP, and further split each layer across 2 GPUs with TP.\nSet dp_shard_size: 4 and tensor_parallel_size: 2.\n\nFSDP + CP on a single 8-GPU node for long context:\n\nYou want to shard the model across all 8 GPUs and also split the sequence length across all 8 GPUs.\nSet dp_shard_size: 8 and context_parallel_size: 8. Note: this means the data parallel group and context parallel group are the same. A more common setup might be to shard across a smaller group.",
"crumbs": [
"Advanced Features",
"N-D Parallelism (Beta)"