Commit Graph

44 Commits

Author SHA1 Message Date
Nicolas Patry a1aac7843b
Choosing input/total tokens automatically based on available VRAM? 2024-10-21 15:06:36 +02:00
OlivierDehaene a6a0c97ed9
feat: prefill chunking (#2600)
* wip

* rollback

* refactor to use prefix/postfix namming + fix all_input_ids_tensor

* maybe patching vlms?

* fix filter and concat

* wip, no filter, no concat

* current

* add prepare_for_prefill

* working

* load tested

* re-create slots

* re-create slots

* fix slot_filtering_indices

* feedback loop

* remove log

* fix benchmarker

* fix vlm and seq2seq

* rename to cache and input lengths

* fix prefill logprobs

* fix launcher

* fix logprobs?

* idk at this point

* max input length

* omfg

* remove debugging lines

* fix tests

* fix mllama

* fix cargo tests

* remove support chunking for paged

* Fixing non blocked attentions

* Fixing dtype + AMD, Ipex targets.

* lint fix.

* rename

* Fix prefix_caching variable, remove defaults in server (confusing a lot
of the times).

* Add simple resolution when user specifies ATTENTION=paged.

* Put back non default simple tests.

* Fix env name

---------

Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>
2024-10-16 12:49:33 +02:00
Nicolas Patry e415b690a6
Lots of improvements (Still 2 allocators) (#2449)
* Making prefix/flashinfer the default and testing the full release tests.

* Include flashinfer in the docker.

* Using prebuilt.

* Allowing window_left_size (dummy version).

* Disabling flashinfer/prefix caching on odd head_dim

* Disable prefix caching for lora.

* More specific codes.

* Update lock

* Updating integration tests with new values with FI/FD.

Remove paged as a default too, and using FD everywhere.

* Update cargo lock ?

* Upgrade to 1.80 because of bitstream...

* Everywhere 1.80

* Forgot last default place.

* Apply suggestions from code review

Co-authored-by: drbh <david.richard.holtz@gmail.com>

* Updated flake lock

* Tmp

* Upgrade resolution system for less errors in resolution.

* Remove lambda for cleaner function.

* Handling debugger.

* OVerride the env in server tests.

* Is this enough to make it work ?

* This seems to be working.

* Downgrade some logs.

* Fixing the default for vlm.

* Don't enable prefix caching on VLM just yet.

* Change `add_special_tokens` in order to have the correct tokens for chat
input and not (since it's super important with the prefixing now)

* Fixing prefix caching for flashdecoding.

* Update all models.

* Fixed flashinfer version.

* add_special_tokens is internal only

* Fixing seqlen with the new vlms.

* Fixing the issue with `add_special_tokens` not being passed around.

* Fixing the test.

* Removing encoder_decoder (seq2seq).

* Update the chat test.

* Fixing the batching tokenization in flash causal lm.

* Truncating left for radix purposes.

* Oops this doesn't belong here.

* Put back default pure shell.

* Update server tests

- Default to throughput test in k6
- Use TGI_WIGGLE_ROOM to adjust wiggle room

* Only n_heads / process_group.size() are necessary.

* Revert the integrationt tests change (seem linked to head_size
modification).

* Adding error message when assert is violated.

* Fixing the free algorithm to handle times where the common prefix is
smaller.

* Apply suggestions from code review

Co-authored-by: OlivierDehaene <olivier@huggingface.co>

* Update server/text_generation_server/layers/attention/common.py

Co-authored-by: OlivierDehaene <olivier@huggingface.co>

* Fix disabling prefix caching - Fix windowing checks.

* Revert the Cohere tokenizer change (for now using a revision instead).

* Fmt.

---------

Co-authored-by: drbh <david.richard.holtz@gmail.com>
Co-authored-by: OlivierDehaene <olivier@huggingface.co>
2024-08-29 16:29:01 +02:00
Daniël de Kok 8deeaca4ff
Add support for prefix caching to the v3 router (#2392)
This change adds support for prefix caching to the v3 router. This
is broken up from the backend support to ease reviewing.

For now prefix caching is only enabled with `USE_PREFIX_CACHING=1`
in this case, the router will switch to `RadixAllocator`. This
allocator uses a radix trie to keep track of prefills that were
seen prior. If a new prefill is a prefix of a previously-seen
prefil, the router will send a request with `prefix_len>0`, which
can be used by the backend to decide to reuse KV blocks from the
cache, rather than recomputing them.

Even though backend support is not added in this PR, the backend
will still work with prefix caching enabled. The prefix lengths
are just ignored and not used.
2024-08-12 14:59:17 +02:00
drbh 04e1af94d7
Enable multiple LoRa adapters (#2010)
* feat: first draft load multiple lora

* feat: load weights within layer and refactor lora pass

* fix: refactor and reduce lora math

* feat: baseline impl single request multi lora support

* feat: prefer lorax implementation and port loading logic

* fix: prefer adapter_data and refactors

* feat: perfer loraxs custom punica kernels and add mlp loras

* fix: adjust batch for bgmv

* fix: adjust adapter_segments logic when in batch

* fix: refactor and move changes to v3 proto

* fix: pass model_id for all flash causal lms

* fix: pass model_id for all causal and seq2seq lms

* fix: add model_id to model test

* feat: add lora support to mistral and refactors

* feat: prefer model id in request

* fix: include rust code for adapter id

* feat: bump launcher and add new lora docs

* feat: support base model generation and refactors

* fix: rename doc to retry ci build

* feat: support if vlm models

* fix: add adapter_data param and avoid missing layers

* fix: add adapter_data param to phi and neox

* fix: update all models forwards to include adapter_data

* fix: add model_id to IdeficsCausalLM

* Update lora.md

Fixed a typo

* Update lora.md

Fixing spam image

* fix: add lora kernel to dockerfile, support running without kernels and refactors

* fix: avoid dockerfile conflict

* fix: refactors and adjust flash llama lora logic

* fix: skip llama test due to CI issue (temp)

* fix: skip llama test CI (temp) 2

* fix: revert skips and prefer updated ci token for tests

* fix: refactors and helpful comments

* fix: add noop in TensorParallelAdapterRowLinear too

* fix: refactor and move shard_lora_weights logic

* fix: exit early if no adapter_data

---------

Co-authored-by: Derek <datavistics@gmail.com>
2024-06-25 14:46:27 -04:00
OlivierDehaene 8aece3bd68
feat: move allocation logic to rust (#1835)
Close #2007
2024-06-05 12:18:38 +02:00
OlivierDehaene 757223b352
feat: add SchedulerV3 (#1996)
- Refactor code to allow supporting multiple versions of the
generate.proto at the same time
- Add v3/generate.proto (ISO to generate.proto for now but allow for
future changes without impacting v2 backends)
- Add Schedule trait to abstract queuing and batching mechanisms that
will be different in the future
- Add SchedulerV2/V3 impl
2024-06-04 15:56:56 +02:00
Daniël de Kok df71aafdcc router: send the input as chunks to the backend
Before this change, the generation input was sent to the backend as a
single string, encoding images as Base64 and packing them in
Markdown-style links.

This change adds a new chunked input representation that separates text
chunks from images chunks. Image chunks contain binary data (for smaller
message sizes) and the image's MIME type.

The stringly-typed inputs are still sent to support backends that do not
support chunked inputs yet.
2024-06-03 17:02:41 +02:00
OlivierDehaene b40e833493
feat: starcoder2 (#1605) 2024-02-28 12:07:08 +01:00
drbh cef0553d59
Outlines guided generation (#1539)
This WIP PR starts to add grammar support via outlines, currently this
PR supports very simple regex grammars and does not optimize for
precompiling or caching grammar fsm's.

todo:
- [X] add simple outlines guidance to `NextTokenChooser`
- [X] update protos for grammar
- [X] update generation params API
- [X] constrain simple grammar
- [ ] support parsing more complex grammar into fsm
- [ ] support all outline support grammar types
- [ ] explore optimizations to avoid recompiling grammars

guided request
```bash
curl -s 'http://localhost:3000/generate' \
--header 'Content-Type: application/json' \
--data-raw '{
    "inputs": "make an email for david: \n",
    "parameters": {
        "max_new_tokens": 6,
        "grammar": "[\\w-]+@([\\w-]+\\.)+[\\w-]+"
    }
}' | jq
```
response
```json
{
  "generated_text": "david@example.com"
}
```

unguided request
```bash
curl -s 'http://localhost:3000/generate' \
--header 'Content-Type: application/json' \
--data '{
    "inputs": "make an email for david: \n",
    "parameters": {
        "max_new_tokens": 6
    }
}' | jq
```
response
```json
{
  "generated_text": "    email = 'david"
}
```
2024-02-15 10:28:10 +01:00
OlivierDehaene 09b7c26bbd
feat(server): add frequency penalty (#1541) 2024-02-08 18:41:25 +01:00
OlivierDehaene d077150eb7
fix: fix gpt-q with groupsize = -1 (#1358) 2023-12-18 16:07:05 +01:00
OlivierDehaene 50b495f3d8
feat: add more latency metrics in forward (#1346) 2023-12-14 15:59:38 +01:00
Nicolas Patry 9ecfa16b12
Speculative (#1308) 2023-12-11 12:46:30 +01:00
OlivierDehaene 3b56d7669b
feat: add mistral model (#1071) 2023-09-28 09:55:47 +02:00
Nicolas Patry 211b54ac41
Rebased #617 (#868)
# What does this PR do?

<!--
Congratulations! You've made it this far! You're not quite done yet
though.

Once merged, your PR is going to appear in the release notes with the
title you set, so make sure it's a great title that fully reflects the
extent of your awesome contribution.

Then, please replace this with a description of the change and which
issue is fixed (if applicable). Please also include relevant motivation
and context. List any dependencies (if any) that are required for this
change.

Once you're done, someone will review your PR shortly (see the section
"Who can review?" below to tag some potential reviewers). They may
suggest changes to make the code even better. If no one reviewed your PR
after a week has passed, don't hesitate to post a new comment
@-mentioning the same persons---sometimes notifications get lost.
-->

<!-- Remove if not applicable -->

Fixes # (issue)


## Before submitting
- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Did you read the [contributor
guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
      Pull Request section?
- [ ] Was this discussed/approved via a Github issue or the
[forum](https://discuss.huggingface.co/)? Please add a link
      to it if that's the case.
- [ ] Did you make sure to update the documentation with your changes?
Here are the
[documentation
guidelines](https://github.com/huggingface/transformers/tree/main/docs),
and
[here are tips on formatting
docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation).
- [ ] Did you write any new necessary tests?


## Who can review?

Anyone in the community is free to review the PR once the tests have
passed. Feel free to tag
members/contributors who may be interested in your PR.

<!-- Your PR will be replied to more quickly if you can figure out the
right person to tag with @


@OlivierDehaene OR @Narsil

 -->

---------

Co-authored-by: Vincent Brouwers <vincent.brouwers@ing.com>
2023-08-28 11:43:47 +02:00
OlivierDehaene fe80f5360c
feat(server): auto max_batch_total_tokens for flash att models (#630) 2023-07-19 09:31:25 +02:00
OlivierDehaene e74bd41e0f
feat(server): add paged attention to flash models (#516)
Closes #478
2023-06-30 19:09:59 +02:00
OlivierDehaene 895c5f1562
feat(server): only compute prefill logprobs when asked (#406)
Close #288
2023-06-02 17:12:30 +02:00
OlivierDehaene 218c9adaa5
feat: decrease IPC proto size (#367)
Closes #307 #308
2023-05-24 19:19:57 +02:00
Nicolas Patry db2b4e0754
feat(router): new healthcheck that skips the queue (#244)
Co-authored-by: OlivierDehaene <23298448+OlivierDehaene@users.noreply.github.com>
Co-authored-by: OlivierDehaene <olivier@huggingface.co>
2023-04-26 20:23:54 +02:00
OlivierDehaene 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
OlivierDehaene 343437c7b5
feat(router): add device and dtype info (#215) 2023-04-21 15:36:29 +02:00
OlivierDehaene 9987960062
feat(router): make router input validation optional (#164) 2023-04-09 20:22:27 +02:00
OlivierDehaene 610bb1f978
feat(benchmark): tui based benchmarking tool (#149) 2023-03-30 15:26:27 +02:00
OlivierDehaene f000068944
feat(server): clear cache on error (#143) 2023-03-28 11:29:35 +02:00
OlivierDehaene b49dbf2d88
fix(server): use server tokenizer as gt (#128) 2023-03-16 12:12:26 +01:00
OlivierDehaene 1a2d68250a
feat: support typical sampling (#114)
closes #112
2023-03-09 11:33:57 +01:00
OlivierDehaene 9b8ea6a6c7
feat(server): add logits watermark (#90) 2023-03-02 12:30:41 +01:00
OlivierDehaene 0ac184ce77
feat(server): add special token bool (#85) 2023-02-24 15:55:57 +01:00
OlivierDehaene 20c3c5940c
feat(router): refactor API and add openAPI schemas (#53) 2023-02-03 12:43:37 +01:00
OlivierDehaene 313194f6d7
feat(server): support repetition penalty (#47) 2023-02-01 15:58:42 +01:00
OlivierDehaene 017a2a8c2f
feat: Add token streaming using ServerSideEvents support (#41) 2023-01-31 17:04:00 +01:00
OlivierDehaene 54fec93193
fix(server): fix seeding with multiple shards (#44) 2023-01-31 16:01:15 +01:00
OlivierDehaene 4f9ac67cfa
Revert "feat: Add token streaming using ServerSideEvents support" (#40)
Reverts huggingface/text-generation-inference#36
2023-01-31 14:21:51 +01:00
OlivierDehaene 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
OlivierDehaene cd298bc5e5
feat: Support sampling seeding (#37)
Co-authored-by: Yannic Kilcher <yk@users.noreply.github.com>
2023-01-30 15:36:16 +01:00
OlivierDehaene 32a253063d
feat: Return logprobs (#8) 2022-12-15 17:03:56 +01:00
OlivierDehaene 718096f695
feat: Support stop sequences (#7) 2022-12-12 18:25:22 +01:00
OlivierDehaene 427d7cc444 feat(server): Support AutoModelForSeq2SeqLM 2022-11-04 18:03:04 +01:00
OlivierDehaene c5665f5c8b feat(server): Support generic AutoModelForCausalLM 2022-11-04 14:22:47 +01:00
Olivier Dehaene f16f2f5ae1 v0.1.0 2022-10-20 19:14:44 +02:00
Olivier Dehaene 4c693e6524 Refactored gRPC interface
Added validation logic
2022-10-11 16:50:54 +02:00
Olivier Dehaene 295831a481 Init 2022-10-08 12:30:12 +02:00