e943a294bc
fix(server): harden the weights choice to save on disk. ( #561 )
...
- Look at `transformers` base class to check for
`_key_to_ignore_on_load_missing` or `_tied_weights` which are the
standard attributes to select the keys to NOT save on disk (since they
are ignored)
- Modified safetensors code (to be reflected in safetensors even if it's
an internal function).
- Will not work for trust_remote_code=True repos (like santacoder).
Should help with :
https://github.com/huggingface/text-generation-inference/issues/555
and : https://github.com/huggingface/text-generation-inference/pull/501
and https://github.com/huggingface/text-generation-inference/issues/556
and
https://github.com/huggingface/text-generation-inference/issues/482#issuecomment-1623713593
2023-07-07 14:50:12 +02:00
abd58ff82c
feat(server): Rework model loading ( #344 )
...
# What does this PR do?
Reworked the loading logic. Idea is to use cleaner loading code:
- Remove need for `no_init_weights`
- Remove all weird `bnb_linear` and `load_weights` and
`post_load_weights`.
New code layout:
- New class `Weights` in charge of handling loading the weights from
multiple files into appropiate tensors (potentially sharded)
- TP layers now are "shells", they contain the code to know what kind of
sharding we need + eventual `all_reduce`. They do not inherit from
linear, but they contain some kind of Linear instead
- the contained linear can be either FastLinear, BnbLinear or GPTq
Linear next.
- All modeling code is explictly made for sharding, process group is
just no-ops for non sharded code (removes a lot of test cases)
---------
Co-authored-by: Ubuntu <ubuntu@ip-172-31-41-161.taildb5d.ts.net>
Co-authored-by: Ubuntu <ubuntu@ip-172-31-41-161.ec2.internal>
Co-authored-by: OlivierDehaene <olivier@huggingface.co>
Co-authored-by: OlivierDehaene <23298448+OlivierDehaene@users.noreply.github.com>
2023-06-08 14:51:52 +02:00
895c5f1562
feat(server): only compute prefill logprobs when asked ( #406 )
...
Close #288
2023-06-02 17:12:30 +02:00
62f91f78ac
feat(server): support vectorized warpers in flash causal lm ( #317 )
...
Co-authored-by: Joel Lamy-Poirier <joel.lamy-poirier@servicenow.com>
2023-05-26 12:30:27 +02:00
218c9adaa5
feat: decrease IPC proto size ( #367 )
...
Closes #307 #308
2023-05-24 19:19:57 +02:00
5a58226130
fix(server): fix decode token ( #334 )
...
Fixes #333
---------
Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>
2023-05-16 23:23:27 +02:00
f092ba9b22
feat(server): add watermarking tests ( #248 )
2023-04-27 19:16:35 +02:00
ebc74d5666
feat(router): use number of tokens in batch as input for dynamic batching ( #226 )
...
Co-authored-by: Nick Hill <nickhill@us.ibm.com>
2023-04-24 17:59:00 +02:00
4a7dd4085a
feat(server): reduce memory requirement ( #214 )
2023-04-24 14:15:42 +02:00
709d8936f6
feat(router): drop requests when client closes the channel ( #202 )
2023-04-20 11:07:40 +02:00
299217c95c
feat(server): add flash attention llama ( #144 )
2023-04-11 16:38:22 +02:00
9987960062
feat(router): make router input validation optional ( #164 )
2023-04-09 20:22:27 +02:00
3f2542bb6a
fix(server): fix escape characters in stop sequence ( #155 )
2023-04-05 19:37:41 +02:00
b49dbf2d88
fix(server): use server tokenizer as gt ( #128 )
2023-03-16 12:12:26 +01:00
1a2d68250a
feat: support typical sampling ( #114 )
...
closes #112
2023-03-09 11:33:57 +01:00
3fef90d50f
feat(clients): Python client ( #103 )
2023-03-07 18:52:22 +01:00
9b205d33cc
fix(server): fix generate_stream by forcing tokens to be decoded correctly ( #100 )
2023-03-06 13:22:58 +01:00
44ce098c10
feat(server): pre-allocate max attention mask ( #75 )
2023-02-24 12:49:21 +01:00
0fbc691946
feat: add safetensors conversion ( #63 )
2023-02-14 13:02:16 +01:00
20c3c5940c
feat(router): refactor API and add openAPI schemas ( #53 )
2023-02-03 12:43:37 +01:00
b1482d9048
breaking(router): modify /generate API to only return generated text ( #50 )
...
@njhill, @yk FYI
generated_text was concatenated to the user prompt for legacy reason. We
want to remove this behaviour as we don't think it is useful and even
detrimonial to usability.
We also remove the unused Vec.
2023-02-02 15:02:04 +01:00
313194f6d7
feat(server): support repetition penalty ( #47 )
2023-02-01 15:58:42 +01:00
f830706b21
feat(server): Support GPT-Neox ( #39 )
2023-01-31 18:53:56 +01:00
017a2a8c2f
feat: Add token streaming using ServerSideEvents support ( #41 )
2023-01-31 17:04:00 +01:00
4f9ac67cfa
Revert "feat: Add token streaming using ServerSideEvents support" ( #40 )
...
Reverts huggingface/text-generation-inference#36
2023-01-31 14:21:51 +01:00
7fbfbb0dc5
feat: Add token streaming using ServerSideEvents support ( #36 )
...
Add token streaming using ServerSideEvents (SSE).
The signature of the SSE events is:
```rust
struct Details {
finish_reason: String,
generated_tokens: u32,
seed: Option<u64>,
}
struct StreamResponse {
token: Token,
generated_text: Option<String>,
details: Option<Details>,
}
struct ErrorResponse {
error: String,
}
```
2023-01-31 11:49:43 +01:00
1f570d181f
fix(server): Fix position ids ( #28 )
2023-01-20 15:35:22 +01:00
15511edc01
feat(server): Support SantaCoder ( #26 )
2023-01-20 12:24:39 +01:00
611e21cb13
fix(server): Fix stop sequences ( #11 )
2022-12-16 16:03:39 +01:00
32a253063d
feat: Return logprobs ( #8 )
2022-12-15 17:03:56 +01:00
718096f695
feat: Support stop sequences ( #7 )
2022-12-12 18:25:22 +01:00
a2985036aa
feat(server): Add model tests ( #6 )
2022-12-08 18:49:33 +01:00
Nicolas Patry