 Alex Cheema
|
9c1bea97e8
fix embed_tokens for last layer in qwen models
|
1 gadu atpakaļ |
|
Alex Cheema
|
af171f06fa
propagate prompts to other nodes so they can display them, cleaner prompt/output output
|
1 gadu atpakaļ |
|
Alex Cheema
|
4a5b80a958
make sure mlx stuff is on separate thread non blocking
|
1 gadu atpakaļ |
|
Alex Cheema
|
6662d5668c
load mlx model shard on mlx thread so it doesnt block
|
1 gadu atpakaļ |
|
Alex Cheema
|
7c649085a1
fix eta/speed for resuming an existing download, using the session downloaded bytes
|
1 gadu atpakaļ |
|
Alex Cheema
|
90e0e2761f
ignore not_started progress updates
|
1 gadu atpakaļ |
|
Alex Cheema
|
265586f7b4
set timeout on get too
|
1 gadu atpakaļ |
|
Alex Cheema
|
4748bb7dc7
increase file download timeout to 30min
|
1 gadu atpakaļ |
|
Alex Cheema
|
ae770db4f3
increase download chunks to 1MB
|
1 gadu atpakaļ |
|
Alex Cheema
|
82f75d0ccf
increase hf download http timeout 15 mins for large downloads
|
1 gadu atpakaļ |
|
Alex Cheema
|
295f41c5cc
increase bench job timeout to give enough time to download
|
1 gadu atpakaļ |
|
Alex Cheema
|
19a27c5bfd
HF_HOME -> EXO_HOME
|
1 gadu atpakaļ |
|
Alex Cheema
|
d7ca9b7732
show each node id in the tinychat topology viz
|
1 gadu atpakaļ |
|
Alex Cheema
|
b349e48b0d
fix visual bug where frontend would show the full hf repo size, but in some cases that includes redundant files so we should use the model index in those cases too
|
1 gadu atpakaļ |
|
Alex Cheema
|
21586063f6
use llama-3.2-1b in tinygrad test
|
1 gadu atpakaļ |
|
Alex Cheema
|
277d63d860
special case when a model doesnt have a model index file, then use wildcard for allow_patterns
|
1 gadu atpakaļ |
|
Alex Cheema
|
74379ef671
log download logs with DEBUG>=6 very verbose
|
1 gadu atpakaļ |
|
Alex Cheema
|
3c7bd48aa3
get rid of some more hf bloat
|
1 gadu atpakaļ |
|
Alex Cheema
|
1df023023e
remove a lot of hf bloat
|
1 gadu atpakaļ |
|
Alex Cheema
|
b89495f444
rewrite ShardDownloader, simplify significantly
|
1 gadu atpakaļ |
|
Alex Cheema
|
a3766f538a
add exception for mlx-community/DeepSeek-R1-3bit and mlx-community/DeepSeek-V3-3bit in tokenizers test
|
1 gadu atpakaļ |
|
Alex Cheema
|
82ef086010
add deepseek-v3-3bit and deepseek-r1-3bit
|
1 gadu atpakaļ |
|
Alex Cheema
|
55ea366932
fix post_init deepseek v3
|
1 gadu atpakaļ |
 Alex Cheema
|
63318983de
Merge pull request #631 from sigseg5/main
|
1 gadu atpakaļ |
 sigseg5
|
fb841a1f50
Adjust truncate size in history list for text without any spaces
|
1 gadu atpakaļ |
|
sigseg5
|
4512366580
Fix bubble behavior when user passes long text without any spaces
|
1 gadu atpakaļ |
|
sigseg5
|
9525c0e7a7
Add adaptive padding for user and assistant messages on width <= 1480px
|
1 gadu atpakaļ |
|
Alex Cheema
|
66f73768cc
Merge pull request #627 from exo-explore/deepseek
|
1 gadu atpakaļ |
|
Alex Cheema
|
fdd05baddb
fix tokenizer tests
|
1 gadu atpakaļ |
|
Alex Cheema
|
59174bdc62
we have a lot of models so group them nicely
|
1 gadu atpakaļ |
