Upload 23 files
Browse files- config/extracted_sample.yaml +40 -0
- config/gpu/compact_openai.yaml +97 -0
- config/gpu/compact_openai_korean.yaml +95 -0
- config/gpu/full_no_rerank_openai.yaml +139 -0
- config/gpu/half_openai.yaml +110 -0
- config/gpu/half_openai_korean.yaml +128 -0
- config/gpu_api/compact_openai.yaml +102 -0
- config/gpu_api/compact_openai_korean.yaml +100 -0
- config/gpu_api/full_no_rerank_openai.yaml +144 -0
- config/gpu_api/half_openai.yaml +115 -0
- config/gpu_api/half_openai_korean.yaml +133 -0
- config/non_gpu/compact_openai.yaml +81 -0
- config/non_gpu/compact_openai_korean.yaml +79 -0
- config/non_gpu/full_no_rerank_openai.yaml +123 -0
- config/non_gpu/half_openai.yaml +94 -0
- config/non_gpu/half_openai_korean.yaml +112 -0
- config/non_gpu/simple_openai.yaml +25 -0
- config/non_gpu/simple_openai_korean.yaml +26 -0
- sample_data/corpus_data_sample.parquet +3 -0
- sample_data/qa_data_sample.parquet +3 -0
- src/__pycache__/runner.cpython-310.pyc +0 -0
- src/runner.py +97 -0
- web.py +326 -0
config/extracted_sample.yaml
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
modules:
|
| 6 |
+
- module_type: vectordb
|
| 7 |
+
embedding_model: openai
|
| 8 |
+
top_k: 3
|
| 9 |
+
strategy:
|
| 10 |
+
metrics:
|
| 11 |
+
- retrieval_f1
|
| 12 |
+
- retrieval_recall
|
| 13 |
+
- retrieval_precision
|
| 14 |
+
- node_line_name: post_retrieve_node_line
|
| 15 |
+
nodes:
|
| 16 |
+
- node_type: prompt_maker
|
| 17 |
+
modules:
|
| 18 |
+
- module_type: fstring
|
| 19 |
+
prompt: "Read the passages and answer the given question. \n Question: {query} \n Passage: {retrieved_contents} \n Answer : "
|
| 20 |
+
strategy:
|
| 21 |
+
generator_modules:
|
| 22 |
+
- batch: 2
|
| 23 |
+
llm: openai
|
| 24 |
+
module_type: llama_index_llm
|
| 25 |
+
metrics:
|
| 26 |
+
- bleu
|
| 27 |
+
- meteor
|
| 28 |
+
- rouge
|
| 29 |
+
- node_type: generator
|
| 30 |
+
modules:
|
| 31 |
+
- batch: 2
|
| 32 |
+
llm: openai
|
| 33 |
+
model: gpt-3.5-turbo-16k
|
| 34 |
+
module_type: llama_index_llm
|
| 35 |
+
strategy:
|
| 36 |
+
metrics:
|
| 37 |
+
- metric_name: bleu
|
| 38 |
+
- metric_name: meteor
|
| 39 |
+
- embedding_model: openai
|
| 40 |
+
metric_name: sem_score
|
config/gpu/compact_openai.yaml
ADDED
|
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: rankgpt
|
| 43 |
+
- module_type: colbert_reranker
|
| 44 |
+
- module_type: sentence_transformer_reranker
|
| 45 |
+
- module_type: flag_embedding_reranker
|
| 46 |
+
- module_type: flag_embedding_llm_reranker
|
| 47 |
+
- module_type: openvino_reranker
|
| 48 |
+
- node_type: passage_filter
|
| 49 |
+
strategy:
|
| 50 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 51 |
+
speed_threshold: 5
|
| 52 |
+
modules:
|
| 53 |
+
- module_type: pass_passage_filter
|
| 54 |
+
- module_type: similarity_threshold_cutoff
|
| 55 |
+
threshold: 0.85
|
| 56 |
+
- module_type: similarity_percentile_cutoff
|
| 57 |
+
percentile: 0.6
|
| 58 |
+
- module_type: threshold_cutoff
|
| 59 |
+
threshold: 0.85
|
| 60 |
+
- module_type: percentile_cutoff
|
| 61 |
+
percentile: 0.6
|
| 62 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 63 |
+
nodes:
|
| 64 |
+
- node_type: prompt_maker
|
| 65 |
+
strategy:
|
| 66 |
+
metrics:
|
| 67 |
+
- metric_name: bleu
|
| 68 |
+
- metric_name: meteor
|
| 69 |
+
- metric_name: rouge
|
| 70 |
+
- metric_name: sem_score
|
| 71 |
+
embedding_model: openai
|
| 72 |
+
speed_threshold: 10
|
| 73 |
+
generator_modules:
|
| 74 |
+
- module_type: llama_index_llm
|
| 75 |
+
llm: openai
|
| 76 |
+
model: [gpt-4o-mini]
|
| 77 |
+
modules:
|
| 78 |
+
- module_type: fstring
|
| 79 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 80 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 81 |
+
- module_type: long_context_reorder
|
| 82 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 83 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 84 |
+
- node_type: generator
|
| 85 |
+
strategy:
|
| 86 |
+
metrics:
|
| 87 |
+
- metric_name: bleu
|
| 88 |
+
- metric_name: meteor
|
| 89 |
+
- metric_name: rouge
|
| 90 |
+
- metric_name: sem_score
|
| 91 |
+
embedding_model: openai
|
| 92 |
+
speed_threshold: 10
|
| 93 |
+
modules:
|
| 94 |
+
- module_type: llama_index_llm
|
| 95 |
+
llm: [openai]
|
| 96 |
+
model: [gpt-4o-mini]
|
| 97 |
+
temperature: [0.5, 1.0]
|
config/gpu/compact_openai_korean.yaml
ADDED
|
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ ko_kiwi ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: rankgpt
|
| 43 |
+
- module_type: colbert_reranker
|
| 44 |
+
- module_type: sentence_transformer_reranker
|
| 45 |
+
- module_type: flag_embedding_reranker
|
| 46 |
+
- module_type: flag_embedding_llm_reranker
|
| 47 |
+
- module_type: openvino_reranker
|
| 48 |
+
- node_type: passage_filter
|
| 49 |
+
strategy:
|
| 50 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 51 |
+
speed_threshold: 5
|
| 52 |
+
modules:
|
| 53 |
+
- module_type: pass_passage_filter
|
| 54 |
+
- module_type: similarity_threshold_cutoff
|
| 55 |
+
threshold: 0.85
|
| 56 |
+
- module_type: similarity_percentile_cutoff
|
| 57 |
+
percentile: 0.6
|
| 58 |
+
- module_type: threshold_cutoff
|
| 59 |
+
threshold: 0.85
|
| 60 |
+
- module_type: percentile_cutoff
|
| 61 |
+
percentile: 0.6
|
| 62 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 63 |
+
nodes:
|
| 64 |
+
- node_type: prompt_maker
|
| 65 |
+
strategy:
|
| 66 |
+
metrics:
|
| 67 |
+
- metric_name: bleu
|
| 68 |
+
- metric_name: meteor
|
| 69 |
+
- metric_name: rouge
|
| 70 |
+
- metric_name: sem_score
|
| 71 |
+
embedding_model: openai
|
| 72 |
+
speed_threshold: 10
|
| 73 |
+
generator_modules:
|
| 74 |
+
- module_type: llama_index_llm
|
| 75 |
+
llm: openai
|
| 76 |
+
model: [gpt-4o-mini]
|
| 77 |
+
modules:
|
| 78 |
+
- module_type: fstring
|
| 79 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 80 |
+
- module_type: long_context_reorder
|
| 81 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 82 |
+
- node_type: generator
|
| 83 |
+
strategy:
|
| 84 |
+
metrics:
|
| 85 |
+
- metric_name: bleu
|
| 86 |
+
- metric_name: meteor
|
| 87 |
+
- metric_name: rouge
|
| 88 |
+
- metric_name: sem_score
|
| 89 |
+
embedding_model: openai
|
| 90 |
+
speed_threshold: 10
|
| 91 |
+
modules:
|
| 92 |
+
- module_type: llama_index_llm
|
| 93 |
+
llm: [openai]
|
| 94 |
+
model: [gpt-4o-mini]
|
| 95 |
+
temperature: [0.5, 1.0]
|
config/gpu/full_no_rerank_openai.yaml
ADDED
|
@@ -0,0 +1,139 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: pre_retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: query_expansion
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 7 |
+
speed_threshold: 10
|
| 8 |
+
top_k: 10
|
| 9 |
+
retrieval_modules:
|
| 10 |
+
- module_type: bm25
|
| 11 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 12 |
+
- module_type: vectordb
|
| 13 |
+
embedding_model: openai
|
| 14 |
+
modules:
|
| 15 |
+
- module_type: pass_query_expansion
|
| 16 |
+
- module_type: query_decompose
|
| 17 |
+
generator_module_type: llama_index_llm
|
| 18 |
+
llm: openai
|
| 19 |
+
model: [ gpt-4o-mini ]
|
| 20 |
+
- module_type: hyde
|
| 21 |
+
generator_module_type: llama_index_llm
|
| 22 |
+
llm: openai
|
| 23 |
+
model: [ gpt-4o-mini ]
|
| 24 |
+
max_token: 64
|
| 25 |
+
- module_type: multi_query_expansion
|
| 26 |
+
generator_module_type: llama_index_llm
|
| 27 |
+
llm: openai
|
| 28 |
+
temperature: [ 0.2, 1.0 ]
|
| 29 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 30 |
+
nodes:
|
| 31 |
+
- node_type: retrieval
|
| 32 |
+
strategy:
|
| 33 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 34 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 10
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: bm25
|
| 39 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 40 |
+
- module_type: vectordb
|
| 41 |
+
embedding_model: openai
|
| 42 |
+
embedding_batch: 256
|
| 43 |
+
- module_type: hybrid_rrf
|
| 44 |
+
weight_range: (4,80)
|
| 45 |
+
- module_type: hybrid_cc
|
| 46 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 47 |
+
weight_range: (0.0, 1.0)
|
| 48 |
+
test_weight_size: 101
|
| 49 |
+
- node_type: passage_augmenter
|
| 50 |
+
strategy:
|
| 51 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 52 |
+
speed_threshold: 5
|
| 53 |
+
top_k: 5
|
| 54 |
+
embedding_model: openai
|
| 55 |
+
modules:
|
| 56 |
+
- module_type: pass_passage_augmenter
|
| 57 |
+
- module_type: prev_next_augmenter
|
| 58 |
+
mode: next
|
| 59 |
+
- node_type: passage_reranker
|
| 60 |
+
strategy:
|
| 61 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 62 |
+
speed_threshold: 10
|
| 63 |
+
top_k: 5
|
| 64 |
+
modules:
|
| 65 |
+
- module_type: pass_reranker
|
| 66 |
+
- module_type: tart
|
| 67 |
+
- module_type: monot5
|
| 68 |
+
- module_type: upr
|
| 69 |
+
- module_type: rankgpt
|
| 70 |
+
- module_type: colbert_reranker
|
| 71 |
+
- module_type: sentence_transformer_reranker
|
| 72 |
+
- module_type: flag_embedding_reranker
|
| 73 |
+
- module_type: flag_embedding_llm_reranker
|
| 74 |
+
- module_type: openvino_reranker
|
| 75 |
+
- node_type: passage_filter
|
| 76 |
+
strategy:
|
| 77 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 78 |
+
speed_threshold: 5
|
| 79 |
+
modules:
|
| 80 |
+
- module_type: pass_passage_filter
|
| 81 |
+
- module_type: similarity_threshold_cutoff
|
| 82 |
+
threshold: 0.85
|
| 83 |
+
- module_type: similarity_percentile_cutoff
|
| 84 |
+
percentile: 0.6
|
| 85 |
+
- module_type: threshold_cutoff
|
| 86 |
+
threshold: 0.85
|
| 87 |
+
- module_type: percentile_cutoff
|
| 88 |
+
percentile: 0.6
|
| 89 |
+
- node_type: passage_compressor
|
| 90 |
+
strategy:
|
| 91 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 92 |
+
speed_threshold: 10
|
| 93 |
+
modules:
|
| 94 |
+
- module_type: pass_compressor
|
| 95 |
+
- module_type: tree_summarize
|
| 96 |
+
llm: openai
|
| 97 |
+
model: gpt-4o-mini
|
| 98 |
+
- module_type: refine
|
| 99 |
+
llm: openai
|
| 100 |
+
model: gpt-4o-mini
|
| 101 |
+
- module_type: longllmlingua
|
| 102 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 103 |
+
nodes:
|
| 104 |
+
- node_type: prompt_maker
|
| 105 |
+
strategy:
|
| 106 |
+
metrics:
|
| 107 |
+
- metric_name: bleu
|
| 108 |
+
- metric_name: meteor
|
| 109 |
+
- metric_name: rouge
|
| 110 |
+
- metric_name: sem_score
|
| 111 |
+
embedding_model: openai
|
| 112 |
+
- metric_name: g_eval
|
| 113 |
+
speed_threshold: 10
|
| 114 |
+
generator_modules:
|
| 115 |
+
- module_type: llama_index_llm
|
| 116 |
+
llm: openai
|
| 117 |
+
model: [gpt-4o-mini]
|
| 118 |
+
modules:
|
| 119 |
+
- module_type: fstring
|
| 120 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 121 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 122 |
+
- module_type: long_context_reorder
|
| 123 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 124 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 125 |
+
- node_type: generator
|
| 126 |
+
strategy:
|
| 127 |
+
metrics:
|
| 128 |
+
- metric_name: bleu
|
| 129 |
+
- metric_name: meteor
|
| 130 |
+
- metric_name: rouge
|
| 131 |
+
- metric_name: sem_score
|
| 132 |
+
embedding_model: openai
|
| 133 |
+
- metric_name: g_eval
|
| 134 |
+
speed_threshold: 10
|
| 135 |
+
modules:
|
| 136 |
+
- module_type: llama_index_llm
|
| 137 |
+
llm: [openai]
|
| 138 |
+
model: [gpt-4o-mini]
|
| 139 |
+
temperature: [0.5, 1.0]
|
config/gpu/half_openai.yaml
ADDED
|
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: rankgpt
|
| 43 |
+
- module_type: colbert_reranker
|
| 44 |
+
- module_type: sentence_transformer_reranker
|
| 45 |
+
- module_type: flag_embedding_reranker
|
| 46 |
+
- module_type: flag_embedding_llm_reranker
|
| 47 |
+
- module_type: openvino_reranker
|
| 48 |
+
- node_type: passage_filter
|
| 49 |
+
strategy:
|
| 50 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 51 |
+
speed_threshold: 5
|
| 52 |
+
modules:
|
| 53 |
+
- module_type: pass_passage_filter
|
| 54 |
+
- module_type: similarity_threshold_cutoff
|
| 55 |
+
threshold: 0.85
|
| 56 |
+
- module_type: similarity_percentile_cutoff
|
| 57 |
+
percentile: 0.6
|
| 58 |
+
- module_type: threshold_cutoff
|
| 59 |
+
threshold: 0.85
|
| 60 |
+
- module_type: percentile_cutoff
|
| 61 |
+
percentile: 0.6
|
| 62 |
+
- node_type: passage_compressor
|
| 63 |
+
strategy:
|
| 64 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 65 |
+
speed_threshold: 10
|
| 66 |
+
modules:
|
| 67 |
+
- module_type: pass_compressor
|
| 68 |
+
- module_type: tree_summarize
|
| 69 |
+
llm: openai
|
| 70 |
+
model: gpt-4o-mini
|
| 71 |
+
- module_type: refine
|
| 72 |
+
llm: openai
|
| 73 |
+
model: gpt-4o-mini
|
| 74 |
+
- module_type: longllmlingua
|
| 75 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 76 |
+
nodes:
|
| 77 |
+
- node_type: prompt_maker
|
| 78 |
+
strategy:
|
| 79 |
+
metrics:
|
| 80 |
+
- metric_name: bleu
|
| 81 |
+
- metric_name: meteor
|
| 82 |
+
- metric_name: rouge
|
| 83 |
+
- metric_name: sem_score
|
| 84 |
+
embedding_model: openai
|
| 85 |
+
speed_threshold: 10
|
| 86 |
+
generator_modules:
|
| 87 |
+
- module_type: llama_index_llm
|
| 88 |
+
llm: openai
|
| 89 |
+
model: [gpt-4o-mini]
|
| 90 |
+
modules:
|
| 91 |
+
- module_type: fstring
|
| 92 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 93 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 94 |
+
- module_type: long_context_reorder
|
| 95 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 96 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 97 |
+
- node_type: generator
|
| 98 |
+
strategy:
|
| 99 |
+
metrics:
|
| 100 |
+
- metric_name: bleu
|
| 101 |
+
- metric_name: meteor
|
| 102 |
+
- metric_name: rouge
|
| 103 |
+
- metric_name: sem_score
|
| 104 |
+
embedding_model: openai
|
| 105 |
+
speed_threshold: 10
|
| 106 |
+
modules:
|
| 107 |
+
- module_type: llama_index_llm
|
| 108 |
+
llm: [openai]
|
| 109 |
+
model: [gpt-4o-mini]
|
| 110 |
+
temperature: [0.5, 1.0]
|
config/gpu/half_openai_korean.yaml
ADDED
|
@@ -0,0 +1,128 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ ko_kiwi ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: rankgpt
|
| 43 |
+
- module_type: colbert_reranker
|
| 44 |
+
- module_type: sentence_transformer_reranker
|
| 45 |
+
- module_type: flag_embedding_reranker
|
| 46 |
+
- module_type: flag_embedding_llm_reranker
|
| 47 |
+
- module_type: openvino_reranker
|
| 48 |
+
- node_type: passage_filter
|
| 49 |
+
strategy:
|
| 50 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 51 |
+
speed_threshold: 5
|
| 52 |
+
modules:
|
| 53 |
+
- module_type: pass_passage_filter
|
| 54 |
+
- module_type: similarity_threshold_cutoff
|
| 55 |
+
threshold: 0.85
|
| 56 |
+
- module_type: similarity_percentile_cutoff
|
| 57 |
+
percentile: 0.6
|
| 58 |
+
- module_type: threshold_cutoff
|
| 59 |
+
threshold: 0.85
|
| 60 |
+
- module_type: percentile_cutoff
|
| 61 |
+
percentile: 0.6
|
| 62 |
+
- node_type: passage_compressor
|
| 63 |
+
strategy:
|
| 64 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 65 |
+
speed_threshold: 10
|
| 66 |
+
modules:
|
| 67 |
+
- module_type: pass_compressor
|
| 68 |
+
- module_type: tree_summarize
|
| 69 |
+
llm: openai
|
| 70 |
+
model: gpt-4o-mini
|
| 71 |
+
prompt: |
|
| 72 |
+
여러 문맥 정보는 다음과 같습니다.\n
|
| 73 |
+
---------------------\n
|
| 74 |
+
{context_str}\n
|
| 75 |
+
---------------------\n
|
| 76 |
+
사전 지식이 아닌 여러 정보가 주어졌습니다,
|
| 77 |
+
질문에 대답하세요.\n
|
| 78 |
+
질문: {query_str}\n
|
| 79 |
+
답변:
|
| 80 |
+
- module_type: refine
|
| 81 |
+
llm: openai
|
| 82 |
+
model: gpt-4o-mini
|
| 83 |
+
prompt: |
|
| 84 |
+
원래 질문은 다음과 같습니다: {query_str}
|
| 85 |
+
기존 답변은 다음과 같습니다: {existing_answer}
|
| 86 |
+
아래에서 기존 답변을 정제할 수 있는 기회가 있습니다.
|
| 87 |
+
(필요한 경우에만) 아래에 몇 가지 맥락을 추가하여 기존 답변을 정제할 수 있습니다.
|
| 88 |
+
------------
|
| 89 |
+
{context_msg}
|
| 90 |
+
------------
|
| 91 |
+
새로운 문맥이 주어지면 기존 답변을 수정하여 질문에 대한 답변을 정제합니다.
|
| 92 |
+
맥락이 쓸모 없다면, 기존 답변을 그대로 답변하세요.
|
| 93 |
+
정제된 답변:
|
| 94 |
+
- module_type: longllmlingua
|
| 95 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 96 |
+
nodes:
|
| 97 |
+
- node_type: prompt_maker
|
| 98 |
+
strategy:
|
| 99 |
+
metrics:
|
| 100 |
+
- metric_name: bleu
|
| 101 |
+
- metric_name: meteor
|
| 102 |
+
- metric_name: rouge
|
| 103 |
+
- metric_name: sem_score
|
| 104 |
+
embedding_model: openai
|
| 105 |
+
speed_threshold: 10
|
| 106 |
+
generator_modules:
|
| 107 |
+
- module_type: llama_index_llm
|
| 108 |
+
llm: openai
|
| 109 |
+
model: [gpt-4o-mini]
|
| 110 |
+
modules:
|
| 111 |
+
- module_type: fstring
|
| 112 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 113 |
+
- module_type: long_context_reorder
|
| 114 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 115 |
+
- node_type: generator
|
| 116 |
+
strategy:
|
| 117 |
+
metrics:
|
| 118 |
+
- metric_name: bleu
|
| 119 |
+
- metric_name: meteor
|
| 120 |
+
- metric_name: rouge
|
| 121 |
+
- metric_name: sem_score
|
| 122 |
+
embedding_model: openai
|
| 123 |
+
speed_threshold: 10
|
| 124 |
+
modules:
|
| 125 |
+
- module_type: llama_index_llm
|
| 126 |
+
llm: [openai]
|
| 127 |
+
model: [gpt-4o-mini]
|
| 128 |
+
temperature: [0.5, 1.0]
|
config/gpu_api/compact_openai.yaml
ADDED
|
@@ -0,0 +1,102 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: cohere_reranker
|
| 43 |
+
- module_type: rankgpt
|
| 44 |
+
- module_type: jina_reranker
|
| 45 |
+
- module_type: colbert_reranker
|
| 46 |
+
- module_type: sentence_transformer_reranker
|
| 47 |
+
- module_type: flag_embedding_reranker
|
| 48 |
+
- module_type: flag_embedding_llm_reranker
|
| 49 |
+
- module_type: time_reranker
|
| 50 |
+
- module_type: openvino_reranker
|
| 51 |
+
- module_type: voyageai_reranker
|
| 52 |
+
- module_type: mixedbreadai_reranker
|
| 53 |
+
- node_type: passage_filter
|
| 54 |
+
strategy:
|
| 55 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 56 |
+
speed_threshold: 5
|
| 57 |
+
modules:
|
| 58 |
+
- module_type: pass_passage_filter
|
| 59 |
+
- module_type: similarity_threshold_cutoff
|
| 60 |
+
threshold: 0.85
|
| 61 |
+
- module_type: similarity_percentile_cutoff
|
| 62 |
+
percentile: 0.6
|
| 63 |
+
- module_type: threshold_cutoff
|
| 64 |
+
threshold: 0.85
|
| 65 |
+
- module_type: percentile_cutoff
|
| 66 |
+
percentile: 0.6
|
| 67 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 68 |
+
nodes:
|
| 69 |
+
- node_type: prompt_maker
|
| 70 |
+
strategy:
|
| 71 |
+
metrics:
|
| 72 |
+
- metric_name: bleu
|
| 73 |
+
- metric_name: meteor
|
| 74 |
+
- metric_name: rouge
|
| 75 |
+
- metric_name: sem_score
|
| 76 |
+
embedding_model: openai
|
| 77 |
+
speed_threshold: 10
|
| 78 |
+
generator_modules:
|
| 79 |
+
- module_type: llama_index_llm
|
| 80 |
+
llm: openai
|
| 81 |
+
model: [gpt-4o-mini]
|
| 82 |
+
modules:
|
| 83 |
+
- module_type: fstring
|
| 84 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 85 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 86 |
+
- module_type: long_context_reorder
|
| 87 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 88 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 89 |
+
- node_type: generator
|
| 90 |
+
strategy:
|
| 91 |
+
metrics:
|
| 92 |
+
- metric_name: bleu
|
| 93 |
+
- metric_name: meteor
|
| 94 |
+
- metric_name: rouge
|
| 95 |
+
- metric_name: sem_score
|
| 96 |
+
embedding_model: openai
|
| 97 |
+
speed_threshold: 10
|
| 98 |
+
modules:
|
| 99 |
+
- module_type: llama_index_llm
|
| 100 |
+
llm: [openai]
|
| 101 |
+
model: [gpt-4o-mini]
|
| 102 |
+
temperature: [0.5, 1.0]
|
config/gpu_api/compact_openai_korean.yaml
ADDED
|
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ ko_kiwi ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: cohere_reranker
|
| 43 |
+
- module_type: rankgpt
|
| 44 |
+
- module_type: jina_reranker
|
| 45 |
+
- module_type: colbert_reranker
|
| 46 |
+
- module_type: sentence_transformer_reranker
|
| 47 |
+
- module_type: flag_embedding_reranker
|
| 48 |
+
- module_type: flag_embedding_llm_reranker
|
| 49 |
+
- module_type: time_reranker
|
| 50 |
+
- module_type: openvino_reranker
|
| 51 |
+
- module_type: voyageai_reranker
|
| 52 |
+
- module_type: mixedbreadai_reranker
|
| 53 |
+
- node_type: passage_filter
|
| 54 |
+
strategy:
|
| 55 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 56 |
+
speed_threshold: 5
|
| 57 |
+
modules:
|
| 58 |
+
- module_type: pass_passage_filter
|
| 59 |
+
- module_type: similarity_threshold_cutoff
|
| 60 |
+
threshold: 0.85
|
| 61 |
+
- module_type: similarity_percentile_cutoff
|
| 62 |
+
percentile: 0.6
|
| 63 |
+
- module_type: threshold_cutoff
|
| 64 |
+
threshold: 0.85
|
| 65 |
+
- module_type: percentile_cutoff
|
| 66 |
+
percentile: 0.6
|
| 67 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 68 |
+
nodes:
|
| 69 |
+
- node_type: prompt_maker
|
| 70 |
+
strategy:
|
| 71 |
+
metrics:
|
| 72 |
+
- metric_name: bleu
|
| 73 |
+
- metric_name: meteor
|
| 74 |
+
- metric_name: rouge
|
| 75 |
+
- metric_name: sem_score
|
| 76 |
+
embedding_model: openai
|
| 77 |
+
speed_threshold: 10
|
| 78 |
+
generator_modules:
|
| 79 |
+
- module_type: llama_index_llm
|
| 80 |
+
llm: openai
|
| 81 |
+
model: [gpt-4o-mini]
|
| 82 |
+
modules:
|
| 83 |
+
- module_type: fstring
|
| 84 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 85 |
+
- module_type: long_context_reorder
|
| 86 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 87 |
+
- node_type: generator
|
| 88 |
+
strategy:
|
| 89 |
+
metrics:
|
| 90 |
+
- metric_name: bleu
|
| 91 |
+
- metric_name: meteor
|
| 92 |
+
- metric_name: rouge
|
| 93 |
+
- metric_name: sem_score
|
| 94 |
+
embedding_model: openai
|
| 95 |
+
speed_threshold: 10
|
| 96 |
+
modules:
|
| 97 |
+
- module_type: llama_index_llm
|
| 98 |
+
llm: [openai]
|
| 99 |
+
model: [gpt-4o-mini]
|
| 100 |
+
temperature: [0.5, 1.0]
|
config/gpu_api/full_no_rerank_openai.yaml
ADDED
|
@@ -0,0 +1,144 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: pre_retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: query_expansion
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 7 |
+
speed_threshold: 10
|
| 8 |
+
top_k: 10
|
| 9 |
+
retrieval_modules:
|
| 10 |
+
- module_type: bm25
|
| 11 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 12 |
+
- module_type: vectordb
|
| 13 |
+
embedding_model: openai
|
| 14 |
+
modules:
|
| 15 |
+
- module_type: pass_query_expansion
|
| 16 |
+
- module_type: query_decompose
|
| 17 |
+
generator_module_type: llama_index_llm
|
| 18 |
+
llm: openai
|
| 19 |
+
model: [ gpt-4o-mini ]
|
| 20 |
+
- module_type: hyde
|
| 21 |
+
generator_module_type: llama_index_llm
|
| 22 |
+
llm: openai
|
| 23 |
+
model: [ gpt-4o-mini ]
|
| 24 |
+
max_token: 64
|
| 25 |
+
- module_type: multi_query_expansion
|
| 26 |
+
generator_module_type: llama_index_llm
|
| 27 |
+
llm: openai
|
| 28 |
+
temperature: [ 0.2, 1.0 ]
|
| 29 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 30 |
+
nodes:
|
| 31 |
+
- node_type: retrieval
|
| 32 |
+
strategy:
|
| 33 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 34 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 10
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: bm25
|
| 39 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 40 |
+
- module_type: vectordb
|
| 41 |
+
embedding_model: openai
|
| 42 |
+
embedding_batch: 256
|
| 43 |
+
- module_type: hybrid_rrf
|
| 44 |
+
weight_range: (4,80)
|
| 45 |
+
- module_type: hybrid_cc
|
| 46 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 47 |
+
weight_range: (0.0, 1.0)
|
| 48 |
+
test_weight_size: 101
|
| 49 |
+
- node_type: passage_augmenter
|
| 50 |
+
strategy:
|
| 51 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 52 |
+
speed_threshold: 5
|
| 53 |
+
top_k: 5
|
| 54 |
+
embedding_model: openai
|
| 55 |
+
modules:
|
| 56 |
+
- module_type: pass_passage_augmenter
|
| 57 |
+
- module_type: prev_next_augmenter
|
| 58 |
+
mode: next
|
| 59 |
+
- node_type: passage_reranker
|
| 60 |
+
strategy:
|
| 61 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 62 |
+
speed_threshold: 10
|
| 63 |
+
top_k: 5
|
| 64 |
+
modules:
|
| 65 |
+
- module_type: pass_reranker
|
| 66 |
+
- module_type: tart
|
| 67 |
+
- module_type: monot5
|
| 68 |
+
- module_type: upr
|
| 69 |
+
- module_type: cohere_reranker
|
| 70 |
+
- module_type: rankgpt
|
| 71 |
+
- module_type: jina_reranker
|
| 72 |
+
- module_type: colbert_reranker
|
| 73 |
+
- module_type: sentence_transformer_reranker
|
| 74 |
+
- module_type: flag_embedding_reranker
|
| 75 |
+
- module_type: flag_embedding_llm_reranker
|
| 76 |
+
- module_type: time_reranker
|
| 77 |
+
- module_type: openvino_reranker
|
| 78 |
+
- module_type: voyageai_reranker
|
| 79 |
+
- module_type: mixedbreadai_reranker
|
| 80 |
+
- node_type: passage_filter
|
| 81 |
+
strategy:
|
| 82 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 83 |
+
speed_threshold: 5
|
| 84 |
+
modules:
|
| 85 |
+
- module_type: pass_passage_filter
|
| 86 |
+
- module_type: similarity_threshold_cutoff
|
| 87 |
+
threshold: 0.85
|
| 88 |
+
- module_type: similarity_percentile_cutoff
|
| 89 |
+
percentile: 0.6
|
| 90 |
+
- module_type: threshold_cutoff
|
| 91 |
+
threshold: 0.85
|
| 92 |
+
- module_type: percentile_cutoff
|
| 93 |
+
percentile: 0.6
|
| 94 |
+
- node_type: passage_compressor
|
| 95 |
+
strategy:
|
| 96 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 97 |
+
speed_threshold: 10
|
| 98 |
+
modules:
|
| 99 |
+
- module_type: pass_compressor
|
| 100 |
+
- module_type: tree_summarize
|
| 101 |
+
llm: openai
|
| 102 |
+
model: gpt-4o-mini
|
| 103 |
+
- module_type: refine
|
| 104 |
+
llm: openai
|
| 105 |
+
model: gpt-4o-mini
|
| 106 |
+
- module_type: longllmlingua
|
| 107 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 108 |
+
nodes:
|
| 109 |
+
- node_type: prompt_maker
|
| 110 |
+
strategy:
|
| 111 |
+
metrics:
|
| 112 |
+
- metric_name: bleu
|
| 113 |
+
- metric_name: meteor
|
| 114 |
+
- metric_name: rouge
|
| 115 |
+
- metric_name: sem_score
|
| 116 |
+
embedding_model: openai
|
| 117 |
+
- metric_name: g_eval
|
| 118 |
+
speed_threshold: 10
|
| 119 |
+
generator_modules:
|
| 120 |
+
- module_type: llama_index_llm
|
| 121 |
+
llm: openai
|
| 122 |
+
model: [gpt-4o-mini]
|
| 123 |
+
modules:
|
| 124 |
+
- module_type: fstring
|
| 125 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 126 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 127 |
+
- module_type: long_context_reorder
|
| 128 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 129 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 130 |
+
- node_type: generator
|
| 131 |
+
strategy:
|
| 132 |
+
metrics:
|
| 133 |
+
- metric_name: bleu
|
| 134 |
+
- metric_name: meteor
|
| 135 |
+
- metric_name: rouge
|
| 136 |
+
- metric_name: sem_score
|
| 137 |
+
embedding_model: openai
|
| 138 |
+
- metric_name: g_eval
|
| 139 |
+
speed_threshold: 10
|
| 140 |
+
modules:
|
| 141 |
+
- module_type: llama_index_llm
|
| 142 |
+
llm: [openai]
|
| 143 |
+
model: [gpt-4o-mini]
|
| 144 |
+
temperature: [0.5, 1.0]
|
config/gpu_api/half_openai.yaml
ADDED
|
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: cohere_reranker
|
| 43 |
+
- module_type: rankgpt
|
| 44 |
+
- module_type: jina_reranker
|
| 45 |
+
- module_type: colbert_reranker
|
| 46 |
+
- module_type: sentence_transformer_reranker
|
| 47 |
+
- module_type: flag_embedding_reranker
|
| 48 |
+
- module_type: flag_embedding_llm_reranker
|
| 49 |
+
- module_type: time_reranker
|
| 50 |
+
- module_type: openvino_reranker
|
| 51 |
+
- module_type: voyageai_reranker
|
| 52 |
+
- module_type: mixedbreadai_reranker
|
| 53 |
+
- node_type: passage_filter
|
| 54 |
+
strategy:
|
| 55 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 56 |
+
speed_threshold: 5
|
| 57 |
+
modules:
|
| 58 |
+
- module_type: pass_passage_filter
|
| 59 |
+
- module_type: similarity_threshold_cutoff
|
| 60 |
+
threshold: 0.85
|
| 61 |
+
- module_type: similarity_percentile_cutoff
|
| 62 |
+
percentile: 0.6
|
| 63 |
+
- module_type: threshold_cutoff
|
| 64 |
+
threshold: 0.85
|
| 65 |
+
- module_type: percentile_cutoff
|
| 66 |
+
percentile: 0.6
|
| 67 |
+
- node_type: passage_compressor
|
| 68 |
+
strategy:
|
| 69 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 70 |
+
speed_threshold: 10
|
| 71 |
+
modules:
|
| 72 |
+
- module_type: pass_compressor
|
| 73 |
+
- module_type: tree_summarize
|
| 74 |
+
llm: openai
|
| 75 |
+
model: gpt-4o-mini
|
| 76 |
+
- module_type: refine
|
| 77 |
+
llm: openai
|
| 78 |
+
model: gpt-4o-mini
|
| 79 |
+
- module_type: longllmlingua
|
| 80 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 81 |
+
nodes:
|
| 82 |
+
- node_type: prompt_maker
|
| 83 |
+
strategy:
|
| 84 |
+
metrics:
|
| 85 |
+
- metric_name: bleu
|
| 86 |
+
- metric_name: meteor
|
| 87 |
+
- metric_name: rouge
|
| 88 |
+
- metric_name: sem_score
|
| 89 |
+
embedding_model: openai
|
| 90 |
+
speed_threshold: 10
|
| 91 |
+
generator_modules:
|
| 92 |
+
- module_type: llama_index_llm
|
| 93 |
+
llm: openai
|
| 94 |
+
model: [gpt-4o-mini]
|
| 95 |
+
modules:
|
| 96 |
+
- module_type: fstring
|
| 97 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 98 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 99 |
+
- module_type: long_context_reorder
|
| 100 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 101 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 102 |
+
- node_type: generator
|
| 103 |
+
strategy:
|
| 104 |
+
metrics:
|
| 105 |
+
- metric_name: bleu
|
| 106 |
+
- metric_name: meteor
|
| 107 |
+
- metric_name: rouge
|
| 108 |
+
- metric_name: sem_score
|
| 109 |
+
embedding_model: openai
|
| 110 |
+
speed_threshold: 10
|
| 111 |
+
modules:
|
| 112 |
+
- module_type: llama_index_llm
|
| 113 |
+
llm: [openai]
|
| 114 |
+
model: [gpt-4o-mini]
|
| 115 |
+
temperature: [0.5, 1.0]
|
config/gpu_api/half_openai_korean.yaml
ADDED
|
@@ -0,0 +1,133 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ ko_kiwi ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_reranker
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 5
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: pass_reranker
|
| 39 |
+
- module_type: tart
|
| 40 |
+
- module_type: monot5
|
| 41 |
+
- module_type: upr
|
| 42 |
+
- module_type: cohere_reranker
|
| 43 |
+
- module_type: rankgpt
|
| 44 |
+
- module_type: jina_reranker
|
| 45 |
+
- module_type: colbert_reranker
|
| 46 |
+
- module_type: sentence_transformer_reranker
|
| 47 |
+
- module_type: flag_embedding_reranker
|
| 48 |
+
- module_type: flag_embedding_llm_reranker
|
| 49 |
+
- module_type: time_reranker
|
| 50 |
+
- module_type: openvino_reranker
|
| 51 |
+
- module_type: voyageai_reranker
|
| 52 |
+
- module_type: mixedbreadai_reranker
|
| 53 |
+
- node_type: passage_filter
|
| 54 |
+
strategy:
|
| 55 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 56 |
+
speed_threshold: 5
|
| 57 |
+
modules:
|
| 58 |
+
- module_type: pass_passage_filter
|
| 59 |
+
- module_type: similarity_threshold_cutoff
|
| 60 |
+
threshold: 0.85
|
| 61 |
+
- module_type: similarity_percentile_cutoff
|
| 62 |
+
percentile: 0.6
|
| 63 |
+
- module_type: threshold_cutoff
|
| 64 |
+
threshold: 0.85
|
| 65 |
+
- module_type: percentile_cutoff
|
| 66 |
+
percentile: 0.6
|
| 67 |
+
- node_type: passage_compressor
|
| 68 |
+
strategy:
|
| 69 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 70 |
+
speed_threshold: 10
|
| 71 |
+
modules:
|
| 72 |
+
- module_type: pass_compressor
|
| 73 |
+
- module_type: tree_summarize
|
| 74 |
+
llm: openai
|
| 75 |
+
model: gpt-4o-mini
|
| 76 |
+
prompt: |
|
| 77 |
+
여러 문맥 정보는 다음과 같습니다.\n
|
| 78 |
+
---------------------\n
|
| 79 |
+
{context_str}\n
|
| 80 |
+
---------------------\n
|
| 81 |
+
사전 지식이 아닌 여러 정보가 주어졌습니다,
|
| 82 |
+
질문에 대답하세요.\n
|
| 83 |
+
질문: {query_str}\n
|
| 84 |
+
답변:
|
| 85 |
+
- module_type: refine
|
| 86 |
+
llm: openai
|
| 87 |
+
model: gpt-4o-mini
|
| 88 |
+
prompt: |
|
| 89 |
+
원래 질문은 다음과 같습니다: {query_str}
|
| 90 |
+
기존 답변은 다음과 같습니다: {existing_answer}
|
| 91 |
+
아래에서 기존 답변을 정제할 수 있는 기회가 있습니다.
|
| 92 |
+
(필요한 경우에만) 아래에 몇 가지 맥락을 추가하여 기존 답변을 정제할 수 있습니다.
|
| 93 |
+
------------
|
| 94 |
+
{context_msg}
|
| 95 |
+
------------
|
| 96 |
+
새로운 문맥이 주어지면 기존 답변을 수정하여 질문에 대한 답변을 정제합니다.
|
| 97 |
+
맥락이 쓸모 없다면, 기존 답변을 그대로 답변하세요.
|
| 98 |
+
정제된 답변:
|
| 99 |
+
- module_type: longllmlingua
|
| 100 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 101 |
+
nodes:
|
| 102 |
+
- node_type: prompt_maker
|
| 103 |
+
strategy:
|
| 104 |
+
metrics:
|
| 105 |
+
- metric_name: bleu
|
| 106 |
+
- metric_name: meteor
|
| 107 |
+
- metric_name: rouge
|
| 108 |
+
- metric_name: sem_score
|
| 109 |
+
embedding_model: openai
|
| 110 |
+
speed_threshold: 10
|
| 111 |
+
generator_modules:
|
| 112 |
+
- module_type: llama_index_llm
|
| 113 |
+
llm: openai
|
| 114 |
+
model: [gpt-4o-mini]
|
| 115 |
+
modules:
|
| 116 |
+
- module_type: fstring
|
| 117 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 118 |
+
- module_type: long_context_reorder
|
| 119 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 120 |
+
- node_type: generator
|
| 121 |
+
strategy:
|
| 122 |
+
metrics:
|
| 123 |
+
- metric_name: bleu
|
| 124 |
+
- metric_name: meteor
|
| 125 |
+
- metric_name: rouge
|
| 126 |
+
- metric_name: sem_score
|
| 127 |
+
embedding_model: openai
|
| 128 |
+
speed_threshold: 10
|
| 129 |
+
modules:
|
| 130 |
+
- module_type: llama_index_llm
|
| 131 |
+
llm: [openai]
|
| 132 |
+
model: [gpt-4o-mini]
|
| 133 |
+
temperature: [0.5, 1.0]
|
config/non_gpu/compact_openai.yaml
ADDED
|
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_filter
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 5
|
| 36 |
+
modules:
|
| 37 |
+
- module_type: pass_passage_filter
|
| 38 |
+
- module_type: similarity_threshold_cutoff
|
| 39 |
+
threshold: 0.85
|
| 40 |
+
- module_type: similarity_percentile_cutoff
|
| 41 |
+
percentile: 0.6
|
| 42 |
+
- module_type: threshold_cutoff
|
| 43 |
+
threshold: 0.85
|
| 44 |
+
- module_type: percentile_cutoff
|
| 45 |
+
percentile: 0.6
|
| 46 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 47 |
+
nodes:
|
| 48 |
+
- node_type: prompt_maker
|
| 49 |
+
strategy:
|
| 50 |
+
metrics:
|
| 51 |
+
- metric_name: bleu
|
| 52 |
+
- metric_name: meteor
|
| 53 |
+
- metric_name: rouge
|
| 54 |
+
- metric_name: sem_score
|
| 55 |
+
embedding_model: openai
|
| 56 |
+
speed_threshold: 10
|
| 57 |
+
generator_modules:
|
| 58 |
+
- module_type: llama_index_llm
|
| 59 |
+
llm: openai
|
| 60 |
+
model: [gpt-4o-mini]
|
| 61 |
+
modules:
|
| 62 |
+
- module_type: fstring
|
| 63 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 64 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 65 |
+
- module_type: long_context_reorder
|
| 66 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 67 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 68 |
+
- node_type: generator
|
| 69 |
+
strategy:
|
| 70 |
+
metrics:
|
| 71 |
+
- metric_name: bleu
|
| 72 |
+
- metric_name: meteor
|
| 73 |
+
- metric_name: rouge
|
| 74 |
+
- metric_name: sem_score
|
| 75 |
+
embedding_model: openai
|
| 76 |
+
speed_threshold: 10
|
| 77 |
+
modules:
|
| 78 |
+
- module_type: llama_index_llm
|
| 79 |
+
llm: [openai]
|
| 80 |
+
model: [gpt-4o-mini]
|
| 81 |
+
temperature: [0.5, 1.0]
|
config/non_gpu/compact_openai_korean.yaml
ADDED
|
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ ko_kiwi ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_filter
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 5
|
| 36 |
+
modules:
|
| 37 |
+
- module_type: pass_passage_filter
|
| 38 |
+
- module_type: similarity_threshold_cutoff
|
| 39 |
+
threshold: 0.85
|
| 40 |
+
- module_type: similarity_percentile_cutoff
|
| 41 |
+
percentile: 0.6
|
| 42 |
+
- module_type: threshold_cutoff
|
| 43 |
+
threshold: 0.85
|
| 44 |
+
- module_type: percentile_cutoff
|
| 45 |
+
percentile: 0.6
|
| 46 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 47 |
+
nodes:
|
| 48 |
+
- node_type: prompt_maker
|
| 49 |
+
strategy:
|
| 50 |
+
metrics:
|
| 51 |
+
- metric_name: bleu
|
| 52 |
+
- metric_name: meteor
|
| 53 |
+
- metric_name: rouge
|
| 54 |
+
- metric_name: sem_score
|
| 55 |
+
embedding_model: openai
|
| 56 |
+
speed_threshold: 10
|
| 57 |
+
generator_modules:
|
| 58 |
+
- module_type: llama_index_llm
|
| 59 |
+
llm: openai
|
| 60 |
+
model: [gpt-4o-mini]
|
| 61 |
+
modules:
|
| 62 |
+
- module_type: fstring
|
| 63 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 64 |
+
- module_type: long_context_reorder
|
| 65 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 66 |
+
- node_type: generator
|
| 67 |
+
strategy:
|
| 68 |
+
metrics:
|
| 69 |
+
- metric_name: bleu
|
| 70 |
+
- metric_name: meteor
|
| 71 |
+
- metric_name: rouge
|
| 72 |
+
- metric_name: sem_score
|
| 73 |
+
embedding_model: openai
|
| 74 |
+
speed_threshold: 10
|
| 75 |
+
modules:
|
| 76 |
+
- module_type: llama_index_llm
|
| 77 |
+
llm: [openai]
|
| 78 |
+
model: [gpt-4o-mini]
|
| 79 |
+
temperature: [0.5, 1.0]
|
config/non_gpu/full_no_rerank_openai.yaml
ADDED
|
@@ -0,0 +1,123 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: pre_retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: query_expansion
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 7 |
+
speed_threshold: 10
|
| 8 |
+
top_k: 10
|
| 9 |
+
retrieval_modules:
|
| 10 |
+
- module_type: bm25
|
| 11 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 12 |
+
- module_type: vectordb
|
| 13 |
+
embedding_model: openai
|
| 14 |
+
modules:
|
| 15 |
+
- module_type: pass_query_expansion
|
| 16 |
+
- module_type: query_decompose
|
| 17 |
+
generator_module_type: llama_index_llm
|
| 18 |
+
llm: openai
|
| 19 |
+
model: [ gpt-4o-mini ]
|
| 20 |
+
- module_type: hyde
|
| 21 |
+
generator_module_type: llama_index_llm
|
| 22 |
+
llm: openai
|
| 23 |
+
model: [ gpt-4o-mini ]
|
| 24 |
+
max_token: 64
|
| 25 |
+
- module_type: multi_query_expansion
|
| 26 |
+
generator_module_type: llama_index_llm
|
| 27 |
+
llm: openai
|
| 28 |
+
temperature: [ 0.2, 1.0 ]
|
| 29 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 30 |
+
nodes:
|
| 31 |
+
- node_type: retrieval
|
| 32 |
+
strategy:
|
| 33 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 34 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 35 |
+
speed_threshold: 10
|
| 36 |
+
top_k: 10
|
| 37 |
+
modules:
|
| 38 |
+
- module_type: bm25
|
| 39 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 40 |
+
- module_type: vectordb
|
| 41 |
+
embedding_model: openai
|
| 42 |
+
embedding_batch: 256
|
| 43 |
+
- module_type: hybrid_rrf
|
| 44 |
+
weight_range: (4,80)
|
| 45 |
+
- module_type: hybrid_cc
|
| 46 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 47 |
+
weight_range: (0.0, 1.0)
|
| 48 |
+
test_weight_size: 101
|
| 49 |
+
- node_type: passage_augmenter
|
| 50 |
+
strategy:
|
| 51 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 52 |
+
speed_threshold: 5
|
| 53 |
+
top_k: 5
|
| 54 |
+
embedding_model: openai
|
| 55 |
+
modules:
|
| 56 |
+
- module_type: pass_passage_augmenter
|
| 57 |
+
- module_type: prev_next_augmenter
|
| 58 |
+
mode: next
|
| 59 |
+
- node_type: passage_filter
|
| 60 |
+
strategy:
|
| 61 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 62 |
+
speed_threshold: 5
|
| 63 |
+
modules:
|
| 64 |
+
- module_type: pass_passage_filter
|
| 65 |
+
- module_type: similarity_threshold_cutoff
|
| 66 |
+
threshold: 0.85
|
| 67 |
+
- module_type: similarity_percentile_cutoff
|
| 68 |
+
percentile: 0.6
|
| 69 |
+
- module_type: threshold_cutoff
|
| 70 |
+
threshold: 0.85
|
| 71 |
+
- module_type: percentile_cutoff
|
| 72 |
+
percentile: 0.6
|
| 73 |
+
- node_type: passage_compressor
|
| 74 |
+
strategy:
|
| 75 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 76 |
+
speed_threshold: 10
|
| 77 |
+
modules:
|
| 78 |
+
- module_type: pass_compressor
|
| 79 |
+
- module_type: tree_summarize
|
| 80 |
+
llm: openai
|
| 81 |
+
model: gpt-4o-mini
|
| 82 |
+
- module_type: refine
|
| 83 |
+
llm: openai
|
| 84 |
+
model: gpt-4o-mini
|
| 85 |
+
- module_type: longllmlingua
|
| 86 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 87 |
+
nodes:
|
| 88 |
+
- node_type: prompt_maker
|
| 89 |
+
strategy:
|
| 90 |
+
metrics:
|
| 91 |
+
- metric_name: bleu
|
| 92 |
+
- metric_name: meteor
|
| 93 |
+
- metric_name: rouge
|
| 94 |
+
- metric_name: sem_score
|
| 95 |
+
embedding_model: openai
|
| 96 |
+
- metric_name: g_eval
|
| 97 |
+
speed_threshold: 10
|
| 98 |
+
generator_modules:
|
| 99 |
+
- module_type: llama_index_llm
|
| 100 |
+
llm: openai
|
| 101 |
+
model: [gpt-4o-mini]
|
| 102 |
+
modules:
|
| 103 |
+
- module_type: fstring
|
| 104 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 105 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 106 |
+
- module_type: long_context_reorder
|
| 107 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 108 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 109 |
+
- node_type: generator
|
| 110 |
+
strategy:
|
| 111 |
+
metrics:
|
| 112 |
+
- metric_name: bleu
|
| 113 |
+
- metric_name: meteor
|
| 114 |
+
- metric_name: rouge
|
| 115 |
+
- metric_name: sem_score
|
| 116 |
+
embedding_model: openai
|
| 117 |
+
- metric_name: g_eval
|
| 118 |
+
speed_threshold: 10
|
| 119 |
+
modules:
|
| 120 |
+
- module_type: llama_index_llm
|
| 121 |
+
llm: [openai]
|
| 122 |
+
model: [gpt-4o-mini]
|
| 123 |
+
temperature: [0.5, 1.0]
|
config/non_gpu/half_openai.yaml
ADDED
|
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ porter_stemmer, space, gpt2 ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_filter
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 5
|
| 36 |
+
modules:
|
| 37 |
+
- module_type: pass_passage_filter
|
| 38 |
+
- module_type: similarity_threshold_cutoff
|
| 39 |
+
threshold: 0.85
|
| 40 |
+
- module_type: similarity_percentile_cutoff
|
| 41 |
+
percentile: 0.6
|
| 42 |
+
- module_type: threshold_cutoff
|
| 43 |
+
threshold: 0.85
|
| 44 |
+
- module_type: percentile_cutoff
|
| 45 |
+
percentile: 0.6
|
| 46 |
+
- node_type: passage_compressor
|
| 47 |
+
strategy:
|
| 48 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 49 |
+
speed_threshold: 10
|
| 50 |
+
modules:
|
| 51 |
+
- module_type: pass_compressor
|
| 52 |
+
- module_type: tree_summarize
|
| 53 |
+
llm: openai
|
| 54 |
+
model: gpt-4o-mini
|
| 55 |
+
- module_type: refine
|
| 56 |
+
llm: openai
|
| 57 |
+
model: gpt-4o-mini
|
| 58 |
+
- module_type: longllmlingua
|
| 59 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 60 |
+
nodes:
|
| 61 |
+
- node_type: prompt_maker
|
| 62 |
+
strategy:
|
| 63 |
+
metrics:
|
| 64 |
+
- metric_name: bleu
|
| 65 |
+
- metric_name: meteor
|
| 66 |
+
- metric_name: rouge
|
| 67 |
+
- metric_name: sem_score
|
| 68 |
+
embedding_model: openai
|
| 69 |
+
speed_threshold: 10
|
| 70 |
+
generator_modules:
|
| 71 |
+
- module_type: llama_index_llm
|
| 72 |
+
llm: openai
|
| 73 |
+
model: [gpt-4o-mini]
|
| 74 |
+
modules:
|
| 75 |
+
- module_type: fstring
|
| 76 |
+
prompt: ["Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 77 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?"]
|
| 78 |
+
- module_type: long_context_reorder
|
| 79 |
+
prompt: [ "Tell me something about the question: {query} \n\n {retrieved_contents}",
|
| 80 |
+
"Question: {query} \n Something to read: {retrieved_contents} \n What's your answer?" ]
|
| 81 |
+
- node_type: generator
|
| 82 |
+
strategy:
|
| 83 |
+
metrics:
|
| 84 |
+
- metric_name: bleu
|
| 85 |
+
- metric_name: meteor
|
| 86 |
+
- metric_name: rouge
|
| 87 |
+
- metric_name: sem_score
|
| 88 |
+
embedding_model: openai
|
| 89 |
+
speed_threshold: 10
|
| 90 |
+
modules:
|
| 91 |
+
- module_type: llama_index_llm
|
| 92 |
+
llm: [openai]
|
| 93 |
+
model: [gpt-4o-mini]
|
| 94 |
+
temperature: [0.5, 1.0]
|
config/non_gpu/half_openai_korean.yaml
ADDED
|
@@ -0,0 +1,112 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision,
|
| 7 |
+
retrieval_ndcg, retrieval_map, retrieval_mrr ]
|
| 8 |
+
speed_threshold: 10
|
| 9 |
+
top_k: 10
|
| 10 |
+
modules:
|
| 11 |
+
- module_type: bm25
|
| 12 |
+
bm25_tokenizer: [ ko_kiwi ]
|
| 13 |
+
- module_type: vectordb
|
| 14 |
+
embedding_model: openai
|
| 15 |
+
embedding_batch: 256
|
| 16 |
+
- module_type: hybrid_rrf
|
| 17 |
+
weight_range: (4,80)
|
| 18 |
+
- module_type: hybrid_cc
|
| 19 |
+
normalize_method: [ mm, tmm, z, dbsf ]
|
| 20 |
+
weight_range: (0.0, 1.0)
|
| 21 |
+
test_weight_size: 101
|
| 22 |
+
- node_type: passage_augmenter
|
| 23 |
+
strategy:
|
| 24 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 25 |
+
speed_threshold: 5
|
| 26 |
+
top_k: 5
|
| 27 |
+
embedding_model: openai
|
| 28 |
+
modules:
|
| 29 |
+
- module_type: pass_passage_augmenter
|
| 30 |
+
- module_type: prev_next_augmenter
|
| 31 |
+
mode: next
|
| 32 |
+
- node_type: passage_filter
|
| 33 |
+
strategy:
|
| 34 |
+
metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
|
| 35 |
+
speed_threshold: 5
|
| 36 |
+
modules:
|
| 37 |
+
- module_type: pass_passage_filter
|
| 38 |
+
- module_type: similarity_threshold_cutoff
|
| 39 |
+
threshold: 0.85
|
| 40 |
+
- module_type: similarity_percentile_cutoff
|
| 41 |
+
percentile: 0.6
|
| 42 |
+
- module_type: threshold_cutoff
|
| 43 |
+
threshold: 0.85
|
| 44 |
+
- module_type: percentile_cutoff
|
| 45 |
+
percentile: 0.6
|
| 46 |
+
- node_type: passage_compressor
|
| 47 |
+
strategy:
|
| 48 |
+
metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
|
| 49 |
+
speed_threshold: 10
|
| 50 |
+
modules:
|
| 51 |
+
- module_type: pass_compressor
|
| 52 |
+
- module_type: tree_summarize
|
| 53 |
+
llm: openai
|
| 54 |
+
model: gpt-4o-mini
|
| 55 |
+
prompt: |
|
| 56 |
+
여러 문맥 정보는 다음과 같습니다.\n
|
| 57 |
+
---------------------\n
|
| 58 |
+
{context_str}\n
|
| 59 |
+
---------------------\n
|
| 60 |
+
사전 지식이 아닌 여러 정보가 주어졌습니다,
|
| 61 |
+
질문에 대답하세요.\n
|
| 62 |
+
질문: {query_str}\n
|
| 63 |
+
답변:
|
| 64 |
+
- module_type: refine
|
| 65 |
+
llm: openai
|
| 66 |
+
model: gpt-4o-mini
|
| 67 |
+
prompt: |
|
| 68 |
+
원래 질문은 다음과 같습니다: {query_str}
|
| 69 |
+
기존 답변은 다음과 같습니다: {existing_answer}
|
| 70 |
+
아래에서 기존 답변을 정제할 수 있는 기회가 있습니다.
|
| 71 |
+
(필요한 경우에만) 아래에 몇 가지 맥락을 추가하여 기존 답변을 정제할 수 있습니다.
|
| 72 |
+
------------
|
| 73 |
+
{context_msg}
|
| 74 |
+
------------
|
| 75 |
+
새로운 문맥이 주어지면 기존 답변을 수정하여 질문에 대한 답변을 정제합니다.
|
| 76 |
+
맥락이 쓸모 없다면, 기존 답변을 그대로 답변하세요.
|
| 77 |
+
정제된 답변:
|
| 78 |
+
- module_type: longllmlingua
|
| 79 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 80 |
+
nodes:
|
| 81 |
+
- node_type: prompt_maker
|
| 82 |
+
strategy:
|
| 83 |
+
metrics:
|
| 84 |
+
- metric_name: bleu
|
| 85 |
+
- metric_name: meteor
|
| 86 |
+
- metric_name: rouge
|
| 87 |
+
- metric_name: sem_score
|
| 88 |
+
embedding_model: openai
|
| 89 |
+
speed_threshold: 10
|
| 90 |
+
generator_modules:
|
| 91 |
+
- module_type: llama_index_llm
|
| 92 |
+
llm: openai
|
| 93 |
+
model: [gpt-4o-mini]
|
| 94 |
+
modules:
|
| 95 |
+
- module_type: fstring
|
| 96 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 97 |
+
- module_type: long_context_reorder
|
| 98 |
+
prompt: ["주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"]
|
| 99 |
+
- node_type: generator
|
| 100 |
+
strategy:
|
| 101 |
+
metrics:
|
| 102 |
+
- metric_name: bleu
|
| 103 |
+
- metric_name: meteor
|
| 104 |
+
- metric_name: rouge
|
| 105 |
+
- metric_name: sem_score
|
| 106 |
+
embedding_model: openai
|
| 107 |
+
speed_threshold: 10
|
| 108 |
+
modules:
|
| 109 |
+
- module_type: llama_index_llm
|
| 110 |
+
llm: [openai]
|
| 111 |
+
model: [gpt-4o-mini]
|
| 112 |
+
temperature: [0.5, 1.0]
|
config/non_gpu/simple_openai.yaml
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 7 |
+
top_k: 3
|
| 8 |
+
modules:
|
| 9 |
+
- module_type: vectordb
|
| 10 |
+
embedding_model: openai
|
| 11 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 12 |
+
nodes:
|
| 13 |
+
- node_type: prompt_maker
|
| 14 |
+
strategy:
|
| 15 |
+
metrics: [bleu, meteor, rouge]
|
| 16 |
+
modules:
|
| 17 |
+
- module_type: fstring
|
| 18 |
+
prompt: "Read the passages and answer the given question. \n Question: {query} \n Passage: {retrieved_contents} \n Answer : "
|
| 19 |
+
- node_type: generator
|
| 20 |
+
strategy:
|
| 21 |
+
metrics: [bleu, meteor, rouge]
|
| 22 |
+
modules:
|
| 23 |
+
- module_type: llama_index_llm
|
| 24 |
+
llm: openai
|
| 25 |
+
model: [gpt-4o-mini]
|
config/non_gpu/simple_openai_korean.yaml
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
node_lines:
|
| 2 |
+
- node_line_name: retrieve_node_line # Arbitrary node line name
|
| 3 |
+
nodes:
|
| 4 |
+
- node_type: retrieval
|
| 5 |
+
strategy:
|
| 6 |
+
metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
|
| 7 |
+
top_k: 3
|
| 8 |
+
modules:
|
| 9 |
+
- module_type: vectordb
|
| 10 |
+
embedding_model: openai
|
| 11 |
+
- node_line_name: post_retrieve_node_line # Arbitrary node line name
|
| 12 |
+
nodes:
|
| 13 |
+
- node_type: prompt_maker
|
| 14 |
+
strategy:
|
| 15 |
+
metrics: [bleu, meteor, rouge]
|
| 16 |
+
modules:
|
| 17 |
+
- module_type: fstring
|
| 18 |
+
prompt: "주어진 passage만을 이용하여 question에 따라 답하시오 passage: {retrieved_contents} \n\n Question: {query} \n\n Answer:"
|
| 19 |
+
- node_type: generator
|
| 20 |
+
strategy:
|
| 21 |
+
metrics: [bleu, meteor, rouge]
|
| 22 |
+
modules:
|
| 23 |
+
- module_type: llama_index_llm
|
| 24 |
+
llm: openai
|
| 25 |
+
model: [gpt-4o-mini]
|
| 26 |
+
batch: 2
|
sample_data/corpus_data_sample.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0fe74568301d61265ce87a76fb7b609f0480e018170d6c275f21c382b1fcb4be
|
| 3 |
+
size 111931
|
sample_data/qa_data_sample.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:70fa30e911d6b748f44e768fe593b6227ba77d6461395e36dc9caf3251f86ab8
|
| 3 |
+
size 9928
|
src/__pycache__/runner.cpython-310.pyc
ADDED
|
Binary file (2.83 kB). View file
|
|
|
src/runner.py
ADDED
|
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import uuid
|
| 3 |
+
from typing import List, Dict, Optional
|
| 4 |
+
|
| 5 |
+
import pandas as pd
|
| 6 |
+
from autorag.deploy import GradioRunner
|
| 7 |
+
from autorag.deploy.api import RetrievedPassage
|
| 8 |
+
from autorag.nodes.generator.base import BaseGenerator
|
| 9 |
+
from autorag.utils import fetch_contents
|
| 10 |
+
|
| 11 |
+
empty_retrieved_passage = RetrievedPassage(
|
| 12 |
+
content="", doc_id="", filepath=None, file_page=None, start_idx=None, end_idx=None
|
| 13 |
+
)
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
class GradioStreamRunner(GradioRunner):
|
| 17 |
+
def __init__(self, config: Dict, project_dir: Optional[str] = None):
|
| 18 |
+
super().__init__(config, project_dir)
|
| 19 |
+
|
| 20 |
+
data_dir = os.path.join(project_dir, "data")
|
| 21 |
+
self.corpus_df = pd.read_parquet(
|
| 22 |
+
os.path.join(data_dir, "corpus.parquet"), engine="pyarrow"
|
| 23 |
+
)
|
| 24 |
+
|
| 25 |
+
def stream_run(self, query: str):
|
| 26 |
+
previous_result = pd.DataFrame(
|
| 27 |
+
{
|
| 28 |
+
"qid": str(uuid.uuid4()),
|
| 29 |
+
"query": [query],
|
| 30 |
+
"retrieval_gt": [[]],
|
| 31 |
+
"generation_gt": [""],
|
| 32 |
+
}
|
| 33 |
+
) # pseudo qa data for execution
|
| 34 |
+
|
| 35 |
+
for module_instance, module_param in zip(
|
| 36 |
+
self.module_instances, self.module_params
|
| 37 |
+
):
|
| 38 |
+
if not isinstance(module_instance, BaseGenerator):
|
| 39 |
+
new_result = module_instance.pure(
|
| 40 |
+
previous_result=previous_result, **module_param
|
| 41 |
+
)
|
| 42 |
+
duplicated_columns = previous_result.columns.intersection(
|
| 43 |
+
new_result.columns
|
| 44 |
+
)
|
| 45 |
+
drop_previous_result = previous_result.drop(
|
| 46 |
+
columns=duplicated_columns
|
| 47 |
+
)
|
| 48 |
+
previous_result = pd.concat(
|
| 49 |
+
[drop_previous_result, new_result], axis=1
|
| 50 |
+
)
|
| 51 |
+
else:
|
| 52 |
+
# retrieved_passages = self.extract_retrieve_passage(
|
| 53 |
+
# previous_result
|
| 54 |
+
# )
|
| 55 |
+
# yield "", retrieved_passages
|
| 56 |
+
# Start streaming of the result
|
| 57 |
+
assert len(previous_result) == 1
|
| 58 |
+
prompt: str = previous_result["prompts"].tolist()[0]
|
| 59 |
+
for delta in module_instance.stream(prompt=prompt,
|
| 60 |
+
**module_param):
|
| 61 |
+
yield delta, [empty_retrieved_passage]
|
| 62 |
+
|
| 63 |
+
def extract_retrieve_passage(self, df: pd.DataFrame) -> List[RetrievedPassage]:
|
| 64 |
+
retrieved_ids: List[str] = df["retrieved_ids"].tolist()[0]
|
| 65 |
+
contents = fetch_contents(self.corpus_df, [retrieved_ids])[0]
|
| 66 |
+
if "path" in self.corpus_df.columns:
|
| 67 |
+
paths = fetch_contents(self.corpus_df, [retrieved_ids], column_name="path")[
|
| 68 |
+
0
|
| 69 |
+
]
|
| 70 |
+
else:
|
| 71 |
+
paths = [None] * len(retrieved_ids)
|
| 72 |
+
metadatas = fetch_contents(
|
| 73 |
+
self.corpus_df, [retrieved_ids], column_name="metadata"
|
| 74 |
+
)[0]
|
| 75 |
+
if "start_end_idx" in self.corpus_df.columns:
|
| 76 |
+
start_end_indices = fetch_contents(
|
| 77 |
+
self.corpus_df, [retrieved_ids], column_name="start_end_idx"
|
| 78 |
+
)[0]
|
| 79 |
+
else:
|
| 80 |
+
start_end_indices = [None] * len(retrieved_ids)
|
| 81 |
+
return list(
|
| 82 |
+
map(
|
| 83 |
+
lambda content, doc_id, path, metadata, start_end_idx: RetrievedPassage(
|
| 84 |
+
content=content,
|
| 85 |
+
doc_id=doc_id,
|
| 86 |
+
filepath=path,
|
| 87 |
+
file_page=metadata.get("page", None),
|
| 88 |
+
start_idx=start_end_idx[0] if start_end_idx else None,
|
| 89 |
+
end_idx=start_end_idx[1] if start_end_idx else None,
|
| 90 |
+
),
|
| 91 |
+
contents,
|
| 92 |
+
retrieved_ids,
|
| 93 |
+
paths,
|
| 94 |
+
metadatas,
|
| 95 |
+
start_end_indices,
|
| 96 |
+
)
|
| 97 |
+
)
|
web.py
ADDED
|
@@ -0,0 +1,326 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import pathlib
|
| 3 |
+
|
| 4 |
+
import gradio as gr
|
| 5 |
+
import pandas as pd
|
| 6 |
+
import yaml
|
| 7 |
+
|
| 8 |
+
from autorag.evaluator import Evaluator
|
| 9 |
+
|
| 10 |
+
from src.runner import GradioStreamRunner
|
| 11 |
+
|
| 12 |
+
root_dir = os.path.dirname(os.path.realpath(__file__))
|
| 13 |
+
|
| 14 |
+
# Paths to example files
|
| 15 |
+
config_dir = os.path.join(root_dir, "config")
|
| 16 |
+
|
| 17 |
+
# Non-GPU Examples
|
| 18 |
+
non_gpu = os.path.join(config_dir, "non_gpu")
|
| 19 |
+
simple_openai = os.path.join(non_gpu, "simple_openai.yaml")
|
| 20 |
+
simple_openai_korean = os.path.join(non_gpu, "simple_openai_korean.yaml")
|
| 21 |
+
compact_openai = os.path.join(non_gpu, "compact_openai.yaml")
|
| 22 |
+
compact_openai_korean = os.path.join(non_gpu, "compact_openai_korean.yaml")
|
| 23 |
+
half_openai = os.path.join(non_gpu, "half_openai.yaml")
|
| 24 |
+
half_openai_korean = os.path.join(non_gpu, "half_openai_korean.yaml")
|
| 25 |
+
full_openai = os.path.join(non_gpu, "full_no_rerank_openai.yaml")
|
| 26 |
+
|
| 27 |
+
non_gpu_examples_list = [
|
| 28 |
+
simple_openai, simple_openai_korean, compact_openai, compact_openai_korean, half_openai, half_openai_korean,
|
| 29 |
+
full_openai
|
| 30 |
+
]
|
| 31 |
+
non_gpu_examples = list(map(lambda x: [x], non_gpu_examples_list))
|
| 32 |
+
|
| 33 |
+
# GPU Examples
|
| 34 |
+
gpu = os.path.join(config_dir, "gpu")
|
| 35 |
+
compact_openai_gpu = os.path.join(gpu, "compact_openai.yaml")
|
| 36 |
+
compact_openai_korean_gpu = os.path.join(gpu, "compact_openai_korean.yaml")
|
| 37 |
+
half_openai_gpu = os.path.join(gpu, "half_openai.yaml")
|
| 38 |
+
half_openai_korean_gpu = os.path.join(gpu, "half_openai_korean.yaml")
|
| 39 |
+
full_openai_gpu = os.path.join(gpu, "full_no_rerank_openai.yaml")
|
| 40 |
+
|
| 41 |
+
gpu_examples_list = [
|
| 42 |
+
compact_openai_gpu, compact_openai_korean_gpu, half_openai_gpu, half_openai_korean_gpu, full_openai_gpu
|
| 43 |
+
]
|
| 44 |
+
gpu_examples = list(map(lambda x: [x], gpu_examples_list))
|
| 45 |
+
|
| 46 |
+
# GPU + API
|
| 47 |
+
gpu_api = os.path.join(config_dir, "gpu_api")
|
| 48 |
+
compact_openai_gpu_api = os.path.join(gpu_api, "compact_openai.yaml")
|
| 49 |
+
compact_openai_korean_gpu_api = os.path.join(gpu_api, "compact_openai_korean.yaml")
|
| 50 |
+
half_openai_gpu_api = os.path.join(gpu_api, "half_openai.yaml")
|
| 51 |
+
half_openai_korean_gpu_api = os.path.join(gpu_api, "half_openai_korean.yaml")
|
| 52 |
+
full_openai_gpu_api = os.path.join(gpu_api, "full_no_rerank_openai.yaml")
|
| 53 |
+
|
| 54 |
+
gpu_api_examples_list = [
|
| 55 |
+
compact_openai_gpu_api, compact_openai_korean_gpu_api, half_openai_gpu_api, half_openai_korean_gpu_api,
|
| 56 |
+
full_openai_gpu_api
|
| 57 |
+
]
|
| 58 |
+
gpu_api_examples = list(map(lambda x: [x], gpu_api_examples_list))
|
| 59 |
+
|
| 60 |
+
example_qa_parquet = os.path.join(root_dir, "sample_data", "qa_data_sample.parquet")
|
| 61 |
+
example_corpus_parquet = os.path.join(root_dir, "sample_data", "corpus_data_sample.parquet")
|
| 62 |
+
|
| 63 |
+
|
| 64 |
+
def display_yaml(file):
|
| 65 |
+
if file is None:
|
| 66 |
+
return "No file uploaded"
|
| 67 |
+
with open(file.name, "r") as f:
|
| 68 |
+
content = yaml.safe_load(f)
|
| 69 |
+
return yaml.dump(content, default_flow_style=False)
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
def display_parquet(file):
|
| 73 |
+
if file is None:
|
| 74 |
+
return pd.DataFrame()
|
| 75 |
+
df = pd.read_parquet(file.name)
|
| 76 |
+
return df
|
| 77 |
+
|
| 78 |
+
|
| 79 |
+
def check_files(yaml_file, qa_file, corpus_file):
|
| 80 |
+
if yaml_file is not None and qa_file is not None and corpus_file is not None:
|
| 81 |
+
return gr.update(visible=True)
|
| 82 |
+
return gr.update(visible=False)
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
def run_trial(file, yaml_file, qa_file, corpus_file):
|
| 86 |
+
project_dir = os.path.join(pathlib.PurePath(file.name).parent, "project")
|
| 87 |
+
evaluator = Evaluator(qa_file, corpus_file, project_dir=project_dir)
|
| 88 |
+
|
| 89 |
+
evaluator.start_trial(yaml_file, skip_validation=True)
|
| 90 |
+
return ("❗Trial Completed❗ "
|
| 91 |
+
"Go to Chat Tab to start the conversation")
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
def set_environment_variable(api_name, api_key):
|
| 95 |
+
if api_name and api_key:
|
| 96 |
+
try:
|
| 97 |
+
os.environ[api_name] = api_key
|
| 98 |
+
return "✅ Setting Complete ✅"
|
| 99 |
+
except Exception as e:
|
| 100 |
+
return f"Error setting environment variable: {e}"
|
| 101 |
+
return "API Name or Key is missing"
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
def stream_default(file, history):
|
| 105 |
+
# Default YAML Runner
|
| 106 |
+
yaml_path = os.path.join(config_dir, "extracted_sample.yaml")
|
| 107 |
+
project_dir = os.path.join(
|
| 108 |
+
pathlib.PurePath(file.name).parent, "project"
|
| 109 |
+
)
|
| 110 |
+
default_gradio_runner = GradioStreamRunner.from_yaml(yaml_path, project_dir)
|
| 111 |
+
|
| 112 |
+
history.append({"role": "assistant", "content": ""})
|
| 113 |
+
# Stream responses for the chatbox
|
| 114 |
+
for default_output in default_gradio_runner.stream_run(history[-2]["content"]):
|
| 115 |
+
stream_delta = default_output[0]
|
| 116 |
+
history[-1]["content"] = stream_delta
|
| 117 |
+
yield history
|
| 118 |
+
|
| 119 |
+
|
| 120 |
+
def stream_optimized(file, history):
|
| 121 |
+
# Custom YAML Runner
|
| 122 |
+
trial_dir = os.path.join(pathlib.PurePath(file.name).parent, "project", "0")
|
| 123 |
+
custom_gradio_runner = GradioStreamRunner.from_trial_folder(trial_dir)
|
| 124 |
+
|
| 125 |
+
history.append({"role": "assistant", "content": ""})
|
| 126 |
+
for output in custom_gradio_runner.stream_run(history[-2]["content"]):
|
| 127 |
+
stream_delta = output[0]
|
| 128 |
+
history[-1]["content"] = stream_delta
|
| 129 |
+
yield history
|
| 130 |
+
|
| 131 |
+
|
| 132 |
+
def user(user_message, history: list):
|
| 133 |
+
return "", history + [{"role": "user", "content": user_message}]
|
| 134 |
+
|
| 135 |
+
|
| 136 |
+
with gr.Blocks(theme="earneleh/paris") as demo:
|
| 137 |
+
gr.Markdown("# AutoRAG Trial & Debugging Interface")
|
| 138 |
+
|
| 139 |
+
with gr.Tabs() as tabs:
|
| 140 |
+
with gr.Tab("Environment Variables"):
|
| 141 |
+
gr.Markdown("## Environment Variables")
|
| 142 |
+
with gr.Row(): # Arrange horizontally
|
| 143 |
+
with gr.Column(scale=3):
|
| 144 |
+
api_name = gr.Textbox(
|
| 145 |
+
label="Environment Variable Name",
|
| 146 |
+
type="text",
|
| 147 |
+
placeholder="Enter your Environment Variable Name",
|
| 148 |
+
)
|
| 149 |
+
gr.Examples(examples=[["OPENAI_API_KEY"]], inputs=api_name)
|
| 150 |
+
with gr.Column(scale=7):
|
| 151 |
+
api_key = gr.Textbox(
|
| 152 |
+
label="API Key",
|
| 153 |
+
type="password",
|
| 154 |
+
placeholder="Enter your API Key",
|
| 155 |
+
)
|
| 156 |
+
|
| 157 |
+
set_env_button = gr.Button("Set Environment Variable")
|
| 158 |
+
env_output = gr.Textbox(
|
| 159 |
+
label="Status", interactive=False
|
| 160 |
+
)
|
| 161 |
+
|
| 162 |
+
api_key.submit(
|
| 163 |
+
set_environment_variable, inputs=[api_name, api_key], outputs=env_output
|
| 164 |
+
)
|
| 165 |
+
set_env_button.click(
|
| 166 |
+
set_environment_variable, inputs=[api_name, api_key], outputs=env_output
|
| 167 |
+
)
|
| 168 |
+
|
| 169 |
+
with gr.Tab("File Upload"):
|
| 170 |
+
with gr.Row() as file_upload_row:
|
| 171 |
+
with gr.Column(scale=3):
|
| 172 |
+
yaml_file = gr.File(
|
| 173 |
+
label="Upload YAML File",
|
| 174 |
+
file_count="single",
|
| 175 |
+
)
|
| 176 |
+
make_yaml_button = gr.Button("Make Your Own YAML File",
|
| 177 |
+
link="https://tally.so/r/mBQY5N")
|
| 178 |
+
|
| 179 |
+
with gr.Column(scale=7):
|
| 180 |
+
yaml_content = gr.Textbox(label="YAML File Content")
|
| 181 |
+
gr.Markdown("Here is the Sample YAML File. Just click the file ❗")
|
| 182 |
+
|
| 183 |
+
gr.Markdown("### Non-GPU Examples")
|
| 184 |
+
gr.Examples(examples=non_gpu_examples, inputs=yaml_file)
|
| 185 |
+
|
| 186 |
+
with gr.Row():
|
| 187 |
+
# Section for GPU examples
|
| 188 |
+
with gr.Column():
|
| 189 |
+
gr.Markdown("### GPU Examples")
|
| 190 |
+
gr.Markdown(
|
| 191 |
+
"**⚠️ Warning**: Here are the YAML files containing the modules that use the **local model**.")
|
| 192 |
+
gr.Markdown(
|
| 193 |
+
"Note that if you Run_Trial in a non-GPU environment, **it can take a very long time**.")
|
| 194 |
+
gr.Examples(examples=gpu_examples, inputs=yaml_file)
|
| 195 |
+
make_gpu = gr.Button("Use AutoRAG GPU Feature",
|
| 196 |
+
link="https://tally.so/r/3j7rP6")
|
| 197 |
+
|
| 198 |
+
# Section for GPU + API examples
|
| 199 |
+
with gr.Column():
|
| 200 |
+
gr.Markdown("### GPU + API Examples")
|
| 201 |
+
gr.Markdown(
|
| 202 |
+
"**⚠️ Warning**: Here are the YAML files containing the modules that use the **local model** and **API Based Model**.")
|
| 203 |
+
gr.Markdown("You need to set **JINA_API_KEY**, **COHERE_API_KEY**, **MXBAI_API_KEY** and **VOYAGE_API_KEY** as environment variables to use this feature. ")
|
| 204 |
+
gr.Examples(examples=gpu_api_examples, inputs=yaml_file)
|
| 205 |
+
gpu_api_button = gr.Button("Use AutoRAG API KEY Feature",
|
| 206 |
+
link="https://tally.so/r/waD1Ab")
|
| 207 |
+
|
| 208 |
+
|
| 209 |
+
|
| 210 |
+
with gr.Row() as qa_upload_row:
|
| 211 |
+
with gr.Column(scale=3):
|
| 212 |
+
qa_file = gr.File(
|
| 213 |
+
label="Upload qa.parquet File",
|
| 214 |
+
file_count="single",
|
| 215 |
+
)
|
| 216 |
+
# Add button for QA
|
| 217 |
+
make_qa_button = gr.Button("Make Your Own QA Data",
|
| 218 |
+
link="https://huggingface.co/spaces/AutoRAG/AutoRAG-data-creation")
|
| 219 |
+
|
| 220 |
+
with gr.Column(scale=7):
|
| 221 |
+
qa_content = gr.Dataframe(label="QA Parquet File Content")
|
| 222 |
+
gr.Markdown("Here is the Sample QA File. Just click the file ❗")
|
| 223 |
+
gr.Examples(examples=[[example_qa_parquet]], inputs=qa_file)
|
| 224 |
+
with gr.Row() as corpus_upload_row:
|
| 225 |
+
with gr.Column(scale=3):
|
| 226 |
+
corpus_file = gr.File(
|
| 227 |
+
label="Upload corpus.parquet File",
|
| 228 |
+
file_count="single",
|
| 229 |
+
)
|
| 230 |
+
make_corpus_button = gr.Button("Make Your Own Corpus Data",
|
| 231 |
+
link="https://huggingface.co/spaces/AutoRAG/AutoRAG-data-creation")
|
| 232 |
+
with gr.Column(scale=7):
|
| 233 |
+
corpus_content = gr.Dataframe(label="Corpus Parquet File Content")
|
| 234 |
+
gr.Markdown(
|
| 235 |
+
"Here is the Sample Corpus File. Just click the file ❗"
|
| 236 |
+
)
|
| 237 |
+
gr.Examples(examples=[[example_corpus_parquet]], inputs=corpus_file)
|
| 238 |
+
|
| 239 |
+
run_trial_button = gr.Button("Run Trial", visible=False)
|
| 240 |
+
trial_output = gr.Textbox(label="Trial Output", visible=False)
|
| 241 |
+
|
| 242 |
+
yaml_file.change(display_yaml, inputs=yaml_file, outputs=yaml_content)
|
| 243 |
+
qa_file.change(display_parquet, inputs=qa_file, outputs=qa_content)
|
| 244 |
+
corpus_file.change(
|
| 245 |
+
display_parquet, inputs=corpus_file, outputs=corpus_content
|
| 246 |
+
)
|
| 247 |
+
|
| 248 |
+
yaml_file.change(
|
| 249 |
+
check_files,
|
| 250 |
+
inputs=[yaml_file, qa_file, corpus_file],
|
| 251 |
+
outputs=run_trial_button,
|
| 252 |
+
)
|
| 253 |
+
qa_file.change(
|
| 254 |
+
check_files,
|
| 255 |
+
inputs=[yaml_file, qa_file, corpus_file],
|
| 256 |
+
outputs=run_trial_button,
|
| 257 |
+
)
|
| 258 |
+
corpus_file.change(
|
| 259 |
+
check_files,
|
| 260 |
+
inputs=[yaml_file, qa_file, corpus_file],
|
| 261 |
+
outputs=run_trial_button,
|
| 262 |
+
)
|
| 263 |
+
|
| 264 |
+
run_trial_button.click(
|
| 265 |
+
lambda: (
|
| 266 |
+
gr.update(visible=False),
|
| 267 |
+
gr.update(visible=False),
|
| 268 |
+
gr.update(visible=False),
|
| 269 |
+
gr.update(visible=True),
|
| 270 |
+
),
|
| 271 |
+
outputs=[
|
| 272 |
+
file_upload_row,
|
| 273 |
+
qa_upload_row,
|
| 274 |
+
corpus_upload_row,
|
| 275 |
+
trial_output,
|
| 276 |
+
],
|
| 277 |
+
)
|
| 278 |
+
run_trial_button.click(
|
| 279 |
+
run_trial,
|
| 280 |
+
inputs=[yaml_file, yaml_file, qa_file, corpus_file],
|
| 281 |
+
outputs=trial_output,
|
| 282 |
+
)
|
| 283 |
+
|
| 284 |
+
# New Chat Tab
|
| 285 |
+
with gr.Tab("Chat") as chat_tab:
|
| 286 |
+
gr.Markdown("### Compare Chat Models")
|
| 287 |
+
|
| 288 |
+
question_input = gr.Textbox(
|
| 289 |
+
label="Your Question", placeholder="Type your question here..."
|
| 290 |
+
)
|
| 291 |
+
pseudo_input = gr.Textbox(label="havertz", visible=False)
|
| 292 |
+
|
| 293 |
+
with gr.Row():
|
| 294 |
+
# Left Chatbox (Default YAML)
|
| 295 |
+
with gr.Column():
|
| 296 |
+
gr.Markdown("#### Naive RAG Chat")
|
| 297 |
+
default_chatbox = gr.Chatbot(label="Naive RAG Conversation",type="messages")
|
| 298 |
+
|
| 299 |
+
# Right Chatbox (Custom YAML)
|
| 300 |
+
with gr.Column():
|
| 301 |
+
gr.Markdown("#### Optimized RAG Chat")
|
| 302 |
+
custom_chatbox = gr.Chatbot(label="Optimized RAG Conversation",type="messages")
|
| 303 |
+
|
| 304 |
+
question_input.submit(lambda x: x, inputs=[question_input], outputs=[pseudo_input]).then(
|
| 305 |
+
user, [question_input, default_chatbox], outputs=[question_input, default_chatbox], queue=False
|
| 306 |
+
).then(
|
| 307 |
+
stream_default,
|
| 308 |
+
inputs=[yaml_file, default_chatbox],
|
| 309 |
+
outputs=[default_chatbox],
|
| 310 |
+
)
|
| 311 |
+
|
| 312 |
+
pseudo_input.change(
|
| 313 |
+
user, [pseudo_input, custom_chatbox], outputs=[question_input, custom_chatbox], queue=False).then(
|
| 314 |
+
stream_optimized,
|
| 315 |
+
inputs=[yaml_file, custom_chatbox],
|
| 316 |
+
outputs=[custom_chatbox],
|
| 317 |
+
)
|
| 318 |
+
|
| 319 |
+
|
| 320 |
+
deploy_button = gr.Button("Deploy",
|
| 321 |
+
link="https://tally.so/r/3XM7y4")
|
| 322 |
+
|
| 323 |
+
|
| 324 |
+
if __name__ == "__main__":
|
| 325 |
+
# Run the interface
|
| 326 |
+
demo.launch(share=False, debug=True)
|