Spaces:

Darshankumarr03
/

mini-rl-openenv

Sleeping

App Files Files Community

mini-rl-openenv

Commit History

Apr 12

14:33
feat: upgrade Router from keyword matching to LLM-powered NLU

Darshan03

cd67e61

Apr 11

16:07
feat: fix avg eval, add per-tier reporting, comprehensive comments, update README v2.2.0

Darshan03

5f336f2
14:44
fix: use grader_score from env metadata instead of broken reward-sum calculation

Darshan03

7198a27
14:34
feat: normalize step-wise rewards to strictly (0, 1) range

Darshan03

a8fff3f
14:29
feat: strictly bound task scores to (0, 1) range for RL training stability

Darshan03

470ff2f

Apr 8

03:42
fix: restore sum/max normalization for final score

Darshan03

7360efa
03:30
fix: align medium_double_charge task with actual DB data (ORD-1515 = Neha Gupta, cosmetics)

Darshan03

acc91ff
03:24
revert: restore original rewards and env logic, keep grader normalization in inference only

Darshan03

06a80a2
03:07
fix: gate resolution/efficiency bonuses by grader_score, use grader for final score, robust reward weighting

Darshan03

83dd0f8
02:34
fix(inference): allow scores to hit bounds [0.0, 1.0]

Darshan03

9eed4e0

Apr 7

15:25
fix(inference): add strict [START]/[STEP]/[END] structured output for openenv validator

Darshan03

13ed1c5
14:50
fixed inference.py

Darshan03

03de6a1

Apr 1

08:27
docs: finalize production-grade documentation and technical comments across core files

Darshan03

0bbc2af
08:00
Docs: Added Environment API Specification to README.md

Darshan03

8d6a8d6
07:52
Final: Enhanced Inference UX and Restored Full 15-Task Suite

Darshan03

34e2954
07:47
Docs: Professional GitHub README while preserving Space metadata

Darshan03

1450b7a
07:43
Final: Multi-Agent Fixes, 8-Task Eval, and RLHF removal

Darshan03

5f13f1d
07:31
Fix: Added required Hugging Face Space YAML metadata to README.md

Darshan03

9119052
07:27
Final: Added Chatbot Feedback, Exit Controls, and Professional Documentation

Darshan03

72165f5
07:23
Final: 100% Full Database Restoration (Untruncated)

Darshan03

4fc1997
07:22
Final: Clear and absolute-import restoration of the Environment API

Darshan03

61770a5
07:20
Fix: Resolved Runtime ImportError and restored missing server files (db.py, tools.py)

Darshan03

b0ca376
07:05
Cleanup: Merged redundant server folders into a single root 'server' directory

Darshan03

2fe9257
07:03
Final: Validated and Windows-ready validator script; Step 3 fixes

Darshan03

5225d67
06:49
Fix: safe .get() on intent_rewards when env not reset; OpenEnv spec /step fix

Darshan03

36ef452
06:39
Deploy standalone Environment API to openenv-csa-rl Space

Darshan03

8b0ccc1
06:24
Implement Official Docker-Integrated Inference Flow for OpenEnv

Darshan03

6931637
06:11
Unify ENV_URL and ensure full .env loading across all modules

Darshan03

e49cee3
06:04
Enhance baseline prompt with Retail Support Protocols for higher scores

Darshan03

2dfaf8b
06:00
Implement Rate Limiting retries and robust LLM parsing in baseline

Darshan03

412bdab
05:58
Add python-dotenv for config management and update .env template

Darshan03

cada417
05:52
Allow dynamic validator ENV_URL assignment in baseline

Darshan03

b151b8a

Mar 31

16:28
Add Like/Dislike RLHF buttons and Exit feature

Darshan03

f32c13d
16:22
Fix UI tearing on chat input container

Darshan03

4d99a92
16:08
Implement OpenEnv specifications, inference framework, and NGINX proxy

Darshan03

2af4435
15:02
feat: enforce protocol via conditional grader and increase API concurrency

Darshan03

0c0fbfa
14:56
chore: final openenv compliance (README tags, Docker healthchecks, baseline scores)

Darshan03

cd6f6a2
11:30
feat: implement autonomous agent loop (Auto-Thinking) for single-turn resolutions

Darshan03

c0adfc1
11:25
feat: add complex scenarios, resolution guides, and E2E demo tab

Darshan03

47c5c21
11:20
feat: include environment feedback in session history for end-to-end resolution

Darshan03

b5411f8
11:19
fix: prevent specialist tool_name hallucinations in prompt

Darshan03

e50a7d4
11:16
perf: optimize generation for speed (greedy, shorter prompt, max_tokens=100)

Darshan03

5dc0e5e
11:10
feat: enhance CSA with reasoning, real-world answers, and final reward display

Darshan03

fdac3d5
11:00
fix: add session-based routes for Streamlit dashboard + update client

Darshan03

92bd01d
10:52
fix: update all consumer files for OpenEnv model migration

Darshan03

c1c99a5
10:43
feat: migrate to native OpenEnv spec + TRL environment_factory⁠

Darshan03

5305c11
09:57
Update verification script to include partial decimal tests (0.1, 0.2, etc.)

Darshan03

ced164f
09:55
Enhance PPO RL setup: Reached 10/10 production grade by adding Action Alignment, empty checks, and strict [-5, 5] stabilization clipping

Darshan03

055d449
09:52
Enhance PPO RL setup: Integrated advanced, continuous, and stabilized reward functions for optimal OpenEnv convergence

Darshan03

987e514
09:47
Add grader boundary verification script to assert 0.0-1.0 scoring range

Darshan03

1f1789d

Commit History

Apr 12

feat: upgrade Router from keyword matching to LLM-powered NLU

Apr 11

feat: fix avg eval, add per-tier reporting, comprehensive comments, update README v2.2.0

fix: use grader_score from env metadata instead of broken reward-sum calculation

feat: normalize step-wise rewards to strictly (0, 1) range

feat: strictly bound task scores to (0, 1) range for RL training stability

Apr 8

fix: restore sum/max normalization for final score

fix: align medium_double_charge task with actual DB data (ORD-1515 = Neha Gupta, cosmetics)

revert: restore original rewards and env logic, keep grader normalization in inference only

fix: gate resolution/efficiency bonuses by grader_score, use grader for final score, robust reward weighting

fix(inference): allow scores to hit bounds [0.0, 1.0]

Apr 7

fix(inference): add strict [START]/[STEP]/[END] structured output for openenv validator

fixed inference.py

Apr 1

docs: finalize production-grade documentation and technical comments across core files

Docs: Added Environment API Specification to README.md

Final: Enhanced Inference UX and Restored Full 15-Task Suite

Docs: Professional GitHub README while preserving Space metadata

Final: Multi-Agent Fixes, 8-Task Eval, and RLHF removal

Fix: Added required Hugging Face Space YAML metadata to README.md

Final: Added Chatbot Feedback, Exit Controls, and Professional Documentation

Final: 100% Full Database Restoration (Untruncated)

Final: Clear and absolute-import restoration of the Environment API

Fix: Resolved Runtime ImportError and restored missing server files (db.py, tools.py)

Cleanup: Merged redundant server folders into a single root 'server' directory

Final: Validated and Windows-ready validator script; Step 3 fixes

Fix: safe .get() on intent_rewards when env not reset; OpenEnv spec /step fix

Deploy standalone Environment API to openenv-csa-rl Space

Implement Official Docker-Integrated Inference Flow for OpenEnv

Unify ENV_URL and ensure full .env loading across all modules

Enhance baseline prompt with Retail Support Protocols for higher scores

Implement Rate Limiting retries and robust LLM parsing in baseline

Add python-dotenv for config management and update .env template

Allow dynamic validator ENV_URL assignment in baseline

Mar 31

Add Like/Dislike RLHF buttons and Exit feature

Fix UI tearing on chat input container

Implement OpenEnv specifications, inference framework, and NGINX proxy

feat: enforce protocol via conditional grader and increase API concurrency

chore: final openenv compliance (README tags, Docker healthchecks, baseline scores)

feat: implement autonomous agent loop (Auto-Thinking) for single-turn resolutions

feat: add complex scenarios, resolution guides, and E2E demo tab

feat: include environment feedback in session history for end-to-end resolution

fix: prevent specialist tool_name hallucinations in prompt

perf: optimize generation for speed (greedy, shorter prompt, max_tokens=100)

feat: enhance CSA with reasoning, real-world answers, and final reward display

fix: add session-based routes for Streamlit dashboard + update client

fix: update all consumer files for OpenEnv model migration

feat: migrate to native OpenEnv spec + TRL environment_factory⁠

Update verification script to include partial decimal tests (0.1, 0.2, etc.)

Enhance PPO RL setup: Reached 10/10 production grade by adding Action Alignment, empty checks, and strict [-5, 5] stabilization clipping

Enhance PPO RL setup: Integrated advanced, continuous, and stabilized reward functions for optimal OpenEnv convergence

Add grader boundary verification script to assert 0.0-1.0 scoring range