biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core][Model runner refactoring 1/N] Refactor attn metadata term (#4518)

Browse Source
This commit is contained in:
SangBin Cho
2024-05-04 02:20:12 +09:00
committed by GitHub
parent 2d7bce9cd5
commit 3521ba4f25
27 changed files with 554 additions and 525 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
tests/spec_decode/test_ngram_worker.py
View File

@@ -43,11 +43,13 @@ def test_ngram_algo_correctness_for_single_no_match():
]
proposal_len = 5
final_seq_lens = [len(prompt) + proposal_len for prompt in prompts]
final_prompt_lens = [len(prompt) + proposal_len for prompt in prompts]
ngram_sampler_output_data = create_execute_model_data(
seq_group_metadata_list=create_seq_group_metadata_from_prompts(
prompts, num_gpu_blocks, block_size,
final_seq_lens=final_seq_lens))
prompts,
num_gpu_blocks,
block_size,
final_prompt_lens=final_prompt_lens))
proposals = proposer.get_proposals(
**ngram_sampler_output_data.to_dict(),
@@ -110,11 +112,13 @@ def test_ngram_algo_correctness_for_batches_not_match_all():
]
proposal_len = 5
final_seq_lens = [len(prompt) + proposal_len for prompt in prompts]
final_prompt_lens = [len(prompt) + proposal_len for prompt in prompts]
ngram_sampler_output_data = create_execute_model_data(
seq_group_metadata_list=create_seq_group_metadata_from_prompts(
prompts, num_gpu_blocks, block_size,
final_seq_lens=final_seq_lens))
prompts,
num_gpu_blocks,
block_size,
final_prompt_lens=final_prompt_lens))
proposals = proposer.get_proposals(
**ngram_sampler_output_data.to_dict(),
@@ -180,11 +184,13 @@ def test_ngram_algo_correctness_for_batches_match_all():
]
proposal_len = 5
final_seq_lens = [len(prompt) + proposal_len for prompt in prompts]
final_prompt_lens = [len(prompt) + proposal_len for prompt in prompts]
ngram_sampler_output_data = create_execute_model_data(
seq_group_metadata_list=create_seq_group_metadata_from_prompts(
prompts, num_gpu_blocks, block_size,
final_seq_lens=final_seq_lens))
prompts,
num_gpu_blocks,
block_size,
final_prompt_lens=final_prompt_lens))
proposals = proposer.get_proposals(
**ngram_sampler_output_data.to_dict(),