Repository navigation
Expand file tree
/
Copy path.env.example
More file actions
202 lines (170 loc) · 9.02 KB
/
Copy path.env.example
File metadata and controls
202 lines (170 loc) · 9.02 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
# =============================================================================
# ExtractBench environment variables
# Copy this file to .env and fill in the values for the providers you want to
# run. The CLI auto-loads .env on startup. Only the pipelines you invoke need
# their variables set.
# =============================================================================
# -----------------------------------------------------------------------------
# Extraction pipelines (the ExtractBench leaderboard roster)
# -----------------------------------------------------------------------------
# LlamaExtract (llamaextract_cost_effective, llamaextract_agentic,
# llamaextract_agentic_plus, llamaextract_agentic_granular_bboxes)
LLAMA_CLOUD_API_KEY=
# Optional: override the API base URL (e.g. a custom deployment)
LLAMA_CLOUD_BASE_URL=
# Optional: EU data residency (llamaparse pipelines with use_europe)
LLAMA_CLOUD_EU_API_KEY=
# Optional: pin uploads to a specific project
LLAMA_CLOUD_PROJECT_ID=
# OpenAI one-shot / two-stage baselines (openai_gpt_5_4_extract_*,
# openai_gpt_5_4_nano_extract_*)
OPENAI_API_KEY=
# Gemini one-shot / two-stage baselines (gemini_3_5_flash_extract_*).
# Any one of GOOGLE_GEMINI_API_KEY / GEMINI_API_KEY / GOOGLE_API_KEY works.
GOOGLE_GEMINI_API_KEY=
# Anthropic one-shot / two-stage baselines (anthropic_haiku_4_5_extract_*)
ANTHROPIC_API_KEY=
# GLM-5.3-flash one-shot baseline
# (glm_5_3_flash_extract_oneshot_structured_output_file), served through z.ai's
# OpenAI-compatible chat API.
GLM_ZAI_API_KEY=
# Coding agents driving extraction:
# - claude_code_extract_opus_4_8 shells out to the Claude Code CLI and uses
# ANTHROPIC_API_KEY above (or a logged-in `claude` session).
# - codex_code_extract_gpt_5_4_low / codex_code_extract_gpt_5_5_high shell out
# to the Codex CLI and read CODEX_API_KEY (an OpenAI API key).
CODEX_API_KEY=
# Tencent Hy4 in the pi coding agent (pi_extract_hy4_vision_evidence), served
# through TokenHub's OpenAI-compatible API.
TENCENT_TOKENHUB_API_KEY=
# Fireworks-hosted pipelines: the DeepSeek two-stage baseline
# (deepseek_v4_pro_extract_twostage_*) and the Kimi K3 one-shot vision pipeline
# (kimi_k3_extract_oneshot_structured_output_file). Set either variable to a
# Fireworks API key.
FIREWORKS_API_KEY=
# A deepseek.com key, for the pipeline that calls DeepSeek's own API
# (deepseek_v4_1_flash_extract_oneshot_structured_output_file). It does not work
# with the Fireworks-hosted pipelines above, and vice versa.
DEEPSEEK_API_KEY=
# Commercial extraction APIs
REDUCTO_API_KEY= # reducto_extract, reducto_extract_v4, reducto_deep_extract (also reducto parse)
EXTEND_API_KEY= # extend_extract, extend_extract_max (also extend parse)
PARSEUR_API_KEY= # parseur_extract_v3_vision
LANDING_AI_API_KEY= # landingai_extract, landingai_parse (VISION_AGENT_API_KEY also accepted)
DATALAB_API_KEY= # datalab_parse_accurate_extract_fast / _balanced, datalab_* parse
# Self-hosted open-weight extraction. There are no default endpoints — each
# variable names a deployment you run yourself.
# Example: https://<your-deployment-host>
LIFT_ENDPOINT_URL= # lift_extract (lift SDK /extract endpoint)
NUEXTRACT3_SERVER_URL= # nuextract3_extract (vLLM server hosting NuExtract3)
NUEXTRACT2_SERVER_URL= # nuextract_2_8b_vllm_extract_* (NuExtract-2.0-8B)
INTERNVL35_8B_SERVER_URL= # internvl3_5_8b_vllm_extract_* (InternVL3.5-8B)
INTERNVL35_14B_SERVER_URL= # internvl3_5_14b_vllm_extract_* (InternVL3.5-14B)
MINICPM_V_SERVER_URL= # minicpm_v_4_5_vllm_extract_* (MiniCPM-V-4.5)
QWEN3_5_35B_SERVER_URL= # qwen3_5_35b_a3b_fp8_vllm_extract_* (Qwen3.5-35B-A3B-FP8)
QWEN3_5_9B_SERVER_URL= # qwen3_5_9b_vllm_extract_* (Qwen3.5-9B)
GEMMA4_E4B_SERVER_URL= # gemma4_e4b_vllm_extract_* (Gemma-4-E4B)
GEMMA4_E2B_SERVER_URL= # gemma4_e2b_vllm_extract_* (Gemma-4-E2B)
KIMI_VL_SERVER_URL= # kimi_vl_a3b_vllm_extract_* (Kimi-VL-A3B-Instruct)
GLM_4_6V_SERVER_URL= # glm_4_6v_flash_vllm_extract_* (GLM-4.6V-Flash)
QWEN3_8_27B_SERVER_URL= # qwen3_8_27b_fp8_vllm_extract_* (Qwen3.8-27B-FP8)
QWEN3_8_FLASH_NEXT_SERVER_URL= # qwen3_8_flash_next_fp8_vllm_extract_* (Qwen3.8-Flash-Next-FP8)
QWEN3_5_2B_SERVER_URL= # qwen3_5_2b_vllm_extract_* (Qwen3.5-2B)
QWEN3_5_0_8B_SERVER_URL= # qwen3_5_0_8b_vllm_extract_* (Qwen3.5-0.8B)
GEMMA4_12B_SERVER_URL= # gemma4_12b_vllm_extract_* (Gemma-4-12B)
QWEN3_5_4B_SERVER_URL= # qwen3_5_4b_vllm_extract_* (Qwen3.5-4B)
# qwen3_6_35b_a3b_fp8_vllm_extract_* and gemma4_26b_vllm_extract_* read
# QWEN35_SERVER_URL and GEMMA4_SERVER_URL, the same endpoints as their parse
# pipelines below.
# -----------------------------------------------------------------------------
# Parse / layout pipelines (inherited ParseBench roster; used by the two-stage
# extraction baselines and cross-evaluation)
# -----------------------------------------------------------------------------
# Mistral OCR (mistral_ocr_4, mistral_ocr_4_annotation)
MISTRAL_API_KEY=
# AWS Textract (aws_textract, aws_textract_with_forms, aws_textract_text_only)
AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_REGION=us-east-2
# Azure Document Intelligence (azure_di_read, azure_di_layout)
AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT=
AZURE_DOCUMENT_INTELLIGENCE_KEY=
# Google Document AI (google_docai, google_docai_layout)
GOOGLE_DOCAI_PROJECT_ID=
GOOGLE_DOCAI_LOCATION=us
GOOGLE_DOCAI_PROCESSOR_ID=
GOOGLE_DOCAI_LAYOUT_PROCESSOR_ID=
# Chunkr (chunkr, chunkr_high_res)
CHUNKR_API_KEY=
# Unstructured (unstructured_auto, unstructured_fast, unstructured_hi_res)
UNSTRUCTURED_API_KEY=
UNSTRUCTURED_API_URL=
# OpenInnovation Parser (oi_parser) — hosted /v1/extract API
# Sign up at https://oi-parser.ai/ to create an account and get an API key.
OI_PARSER_API_KEY=
# Optional: override the API base URL (defaults to the service's public endpoint)
OI_PARSER_BASE_URL=
PULSE_API_KEY=
# Optional Pulse tuning: PULSE_API_BASE_URL / PULSE_JOB_TIMEOUT /
# PULSE_POLL_INTERVAL / PULSE_REQUEST_TIMEOUT
# Databricks ai_parse (databricks_ai_parse, databricks_ai_parse_batch)
DATABRICKS_HOST=
DATABRICKS_TOKEN=
DATABRICKS_SQL_WAREHOUSE_ID=
DATABRICKS_AI_PARSE_VOLUME=
# dots.ocr hosted endpoint (dots_ocr_1_0_parse, dots_ocr_1_5_parse)
DOTS_OCR_ENDPOINT_URL=
DOTS_OCR_API_KEY=
# KDL frontier nano (kdl_frontier_nano)
KDL_NANO_ENDPOINT_URL=
# -----------------------------------------------------------------------------
# Self-hosted model servers (required by the matching pipelines; no defaults)
# -----------------------------------------------------------------------------
# Layout detection
SURYA_ENDPOINT_URL= # surya_layout
CHANDRA_ENDPOINT_URL= # chandra_layout (OpenAI-compatible /v1 base URL)
QWEN3VL_ENDPOINT_URL= # qwen3vl_layout, qwen3vl_parse (/v1 base URL)
LAYOUT_V3_BYOC_CPU_URL= # layout_v3_byoc_cpu
LAYOUT_V3_BYOC_GPU_URL= # layout_v3_byoc_gpu
# Layout detection served behind a HuggingFace-Inference-Endpoint-style API
# (these also need HF_TOKEN below). Deploy the model yourself and point here.
LAYOUT_V3_ENDPOINT_URL= # layout_v3
YOLO_LAYOUT_ENDPOINT_URL= # yolo_doclaynet
PADDLE_LAYOUT_ENDPOINT_URL= # ppdoclayout_plus_l
DOCLING_LAYOUT_ENDPOINT_URL= # docling_layout_old
DOCLING_LAYOUT_HERON_ENDPOINT_URL= # docling_layout_heron
DOCLING_LAYOUT_HERON_101_ENDPOINT_URL= # docling_layout_heron_101
# vLLM / SDK-served parse models
CHANDRA2_SERVER_URL= # chandra2_vllm, chandra2_sdk
SURYA2_SERVER_URL= # surya2_sdk
MINERU25_SERVER_URL= # mineru25_vllm
MINERU2605PRO_SERVER_URL= # mineru2605pro_vllm
PADDLEOCR_SERVER_URL= # paddleocr_vl_*_vllm / _pipeline
GEMMA4_SERVER_URL= # gemma4_*_vllm
QWEN35_SERVER_URL= # qwen3_5_*/qwen3_6_*_vllm_parse and _vllm_layout
DEEPSEEKOCR2_SERVER_URL= # deepseekocr2_vllm, deepseekocr2_freeocr
GRANITE_VISION_SERVER_URL= # granite_vision_pipeline, granite_vision_4_1_4b
UNLIMITEDOCR_SERVER_URL= # unlimitedocr
MINERU_DIFFUSION_SERVER_URL= # mineru_diffusion
NEMOTRON_OMNI_SERVER_URL= # nemotron_omni_30b_vllm_thinking
DOCLING_PARSE_ENDPOINT_URL= # docling_parse
DOCLING_PARSE_API_KEY=
DOCLING_SERVE_ENDPOINT_URL= # docling_serve
DOCLING_SERVE_API_KEY=
# VLLM_API_KEY= # only if your vLLM servers require auth
# HuggingFace token: required by layout_v3* checkpoints and any gated models
HF_TOKEN=
# -----------------------------------------------------------------------------
# Library-only providers (importable, but no registered pipeline yet)
# -----------------------------------------------------------------------------
# GLM_API_KEY=
# AZURE_CONTENT_UNDERSTANDING_ENDPOINT=
# AZURE_CONTENT_UNDERSTANDING_KEY=
# -----------------------------------------------------------------------------
# Evaluation options
# -----------------------------------------------------------------------------
# Set to 0 to force the reference APTED implementation for TEDS
# EXTRACT_BENCH_FAST_TEDS=1
# Gemini CLI uses the Google credentials above. Optional native Gemini gateway:
# GOOGLE_GEMINI_BASE_URL=https://your-gateway.example/genai
# GEMINI_CLI_CUSTOM_HEADERS=x-model-provider: gemini