Port fixes from master to 2024.5.1 / 2024.6.0 #1239
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Always tokenize as batch to return `attention_mask` to preserve the error: ``` The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results. ```
…oolkit#1189) See https://productionresultssa7.blob.core.windows.net/actions-results/1d9c7ed7-f60f-463a-8dba-2898f6f39e0a/workflow-job-run-2ad6080a-b02b-5d2d-feb5-675d9367578e/logs/job/job-logs.txt?rsct=text%2Fplain&se=2024-11-11T12%3A35%3A03Z&sig=cE1uxQm2H0TswczNt4jESUwM1%2BzN2uQBHxsm7Tqw0AI%3D&ske=2024-11-11T22%3A38%3A18Z&skoid=ca7593d4-ee42-46cd-af88-8b886a2f84eb&sks=b&skt=2024-11-11T10%3A38%3A18Z&sktid=398a6654-997b-47e9-b12b-9515b896b4de&skv=2024-08-04&sp=r&spr=https&sr=b&st=2024-11-11T12%3A24%3A58Z&sv=2024-08-04 A lot of words are printed as max_new_tokens is quite huge number, while exception is expected as streamer cannot work with batch. To fix hanging of tests in this PR openvinotoolkit#882
) Fixed ``` model_tmp_path = ('katuni4ka/tiny-random-phi3', WindowsPath('C:/Users/runneradmin/AppData/Local/Temp/pytest-of-runneradmin/pytest-0/katuni4ka_tiny-random-phi31')) generation_config = {'eos_token_id': 42, 'ignore_eos': True} @pytest.mark.precommit @pytest.mark.nightly @pytest.mark.parametrize("generation_config", invalid_py_configs) def test_python_generation_config_validation(model_tmp_path, generation_config): model_id, temp_path = model_tmp_path pipe = load_pipe([({"eos_token_id": 37}, "config.json")], temp_path) # 'unexisting_key_name' key validity is checked in pybind and ValueError will be returned # instead of RuntimeError, which is returned when GenerationConfig values are validated return_exception_type = ValueError if 'unexisting_key_name' in generation_config else RuntimeError > with pytest.raises(return_exception_type): E Failed: DID NOT RAISE <class 'RuntimeError'> ```
Fixed `test_valid_configs`: ``` FAILED tests/python_tests/test_generate_api.py::test_valid_configs - RuntimeError: Check 'eos_token_id != -1 || max_new_tokens != (18446744073709551615UL) || max_length != (18446744073709551615UL)' failed at /home/runner/work/openvino.genai/openvino.genai/src/cpp/src/generation_config.cpp:164: Either 'eos_token_id', or 'max_new_tokens', or 'max_length' should be defined. ``` See https://github.com/openvinotoolkit/openvino.genai/actions/runs/11797566325/job/32862069017?pr=882
External chat template is not patched for some reason, but it still can contain JinjaCpp unsupported constructions.
Added a command for exporting compressed VLMs similar to how it is done for LMs.
Task: [CVS-155520](https://jira.devtools.intel.com/browse/CVS-155520) --------- Co-authored-by: Ekaterina Aidova <[email protected]>
…penvinotoolkit#1222) When using NPU, it seems that the eos token is not initialized correctly (at least, for certain models). This causes chat sample to have a conversation with itself: ``` >chat_sample.exe Meta-Llama-3-8B-Instruct question: hello! Hello! It's nice to meet you! Is there something I can help you with, or would you like to chat?assistant Nice to meet you too! I'm just a language model, I don't have personal experiences or emotions, but I'm here to help answer any questions you might have or engage in a fun conversation! What's on your mind? Want to talk about something in particular or just shoot the breeze?assistant Sounds like fun! I ---------- question: ``` Borrowing some initialization code from *StatefulLLMPipeline*, where we init eos token from tokenizer within constructor, if eos token has not been provided, the issue is resolved: ``` > chat_sample.exe Meta-Llama-3-8B-Instruct question: hello! Hello! It's nice to meet you! Is there something I can help you with, or would you like to chat? ---------- question: ```
…antization (openvinotoolkit#1228) Added a command for exporting quantized diffusion models similar to how it is done for LMs.
github-actions
bot
added
category: llm_bench
ilya-lavrenov
force-pushed
the
port-fixes
branch
3 times, most recently
from
3013560
to
4aa9ed7
Compare
ilya-lavrenov
force-pushed
the
port-fixes
branch
from
4aa9ed7
to
2f10139
Compare
|
@AlexKoff88 please, suggest a fix that needs to be ported to resolve https://github.com/openvinotoolkit/openvino.genai/actions/runs/11939760566/job/33280974703?pr=1239 |
The tests will be fixed in #1238. I am ok with merging this PR. |
AlexKoff88
approved these changes
Nov 21, 2024
github-merge-queue
bot
removed this pull request from the merge queue due to failed status checks
Merged
via the queue into
openvinotoolkit:releases/2024/5
with commit Nov 21, 2024
da7a7ca
52 of 54 checks passed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Ported: