Alex Cheema
|
9345684b38
closely match prev impl mlx non blocking
|
11 mēneši atpakaļ |
Alex Cheema
|
d6e661fd69
match previous impl with np.array in mlx
|
11 mēneši atpakaļ |
Alex Cheema
|
caf9b57a2a
trigger ci
|
11 mēneši atpakaļ |
Alex Cheema
|
84187113de
add a test for hf get_weight_map
|
11 mēneši atpakaļ |
Alex Cheema
|
4ec613d4e8
simplify tinygrad non blocking
|
11 mēneši atpakaļ |
Alex Cheema
|
6342384df4
Merge branch 'main' into non_blocking
|
11 mēneši atpakaļ |
Alex Cheema
|
a1a0ffac55
add tinychat option for llama-3.1-70b-bf16
|
11 mēneši atpakaļ |
Alex Cheema
|
de19f0ab42
Merge branch 'main' into non_blocking
|
11 mēneši atpakaļ |
Alex Cheema
|
2948a83448
add llama-3.1-70b-bf16 model option
|
11 mēneši atpakaļ |
Alex Cheema
|
11dd952d26
use set for shard specific patterns
|
11 mēneši atpakaļ |
Alex Cheema
|
ea3322dea4
remove comment
|
11 mēneši atpakaļ |
Alex Cheema
|
e0fda94d20
use sets for shard specific patterns
|
11 mēneši atpakaļ |
Alex Cheema
|
b239c8a6d0
Merge branch 'main' into non_blocking
|
11 mēneši atpakaļ |
Alex Cheema
|
8f65e1e697
fix weight_map resolution. previously we were always defaulting to allow pattern *.safetensors
|
11 mēneši atpakaļ |
Alex Cheema
|
6881722b72
simplify non-blocking mlx inference
|
11 mēneši atpakaļ |
Alex Cheema
|
9db16f8dca
use a queue for non-blocking mlx inference
|
11 mēneši atpakaļ |
Alex Cheema
|
0ca5c26094
run mlx inference engine on a single thread too
|
11 mēneši atpakaļ |
Alex Cheema
|
58f535d0b0
formatting
|
11 mēneši atpakaļ |
Alex Cheema
|
2950373d36
experiment with tinygrad on its own thread, so it doesnt block event loop
|
11 mēneši atpakaļ |
Alex Cheema
|
41f0a22e76
DEBUG>=8 for SendOpaqueStatus logs
|
11 mēneši atpakaļ |
Alex Cheema
|
01cc6a4c9d
fix Mistral-Large special case when we pass in a path
|
11 mēneši atpakaļ |
Alex Cheema
|
41dd700ff4
less aggressive logs for opaque status / download progress. too much spam
|
11 mēneši atpakaļ |
Alex Cheema
|
35aba75be6
Merge pull request #194 from exo-explore/better_networking
|
11 mēneši atpakaļ |
Alex Cheema
|
4537d614aa
circleci use tee to output logs in realtime as well as capture them
|
11 mēneši atpakaļ |
Alex Cheema
|
56c1bf9a95
consistent remove _secs / -secs suffix
|
11 mēneši atpakaļ |
Alex Cheema
|
f342cdcae9
get rid of -secs suffix
|
11 mēneši atpakaļ |
Alex Cheema
|
a0d9c90e96
shorten cli name --chatgpt-api-response-timeout
|
11 mēneši atpakaļ |
Alex Cheema
|
8cb678e795
better logs around peer connecting / disconnecting
|
11 mēneši atpakaļ |
Alex Cheema
|
c97da5480f
add id to set
|
11 mēneši atpakaļ |
Alex Cheema
|
80c48b9e76
update visited with self.id, timeout on collecting topology from a peer 5s
|
11 mēneši atpakaļ |