Spaces:

DeepikaChintamreddy
/

config-debug-env

Sleeping

App Files Files Community

config-debug-env / server

Commit History

Apr 12

18:16
fix: eliminate all 0.0 reward values across all files

Deepikachintamreddy

bc98e17
18:09
fix: progress_ratio default 0.01

Deepikachintamreddy

b1c5449
18:05
fix: total_reward init 0.01

Deepikachintamreddy

3ca9f73
17:16
fix: remove function aliases from grader_api, classes only, NaN protection

Deepikachintamreddy

d500ed3
16:37
fix: grader classes with grade() method per Meta feedback

Deepikachintamreddy

b5751d2
16:24
fix: model defaults 0.01 not 0.0

Deepikachintamreddy

c8197e5
14:45
fix: initial previous_reward 0.01 not 0.0

Deepikachintamreddy

991b3a8
14:42
fix: reset reward 0.01 not 0.0

Deepikachintamreddy

f198438
14:35
fix: grader handles all call patterns, reset reward 0.01

Deepikachintamreddy

b17299b
14:10
fix: class-based graders with grade() method per Meta instructions

Deepikachintamreddy

b5fc338
13:40
fix: bulletproof graders - never throw, always return float in range

Deepikachintamreddy

5b5379b
13:00
fix: grader_api returns float only, not tuple

Deepikachintamreddy

e421a0a
12:28
fix: correct step format, per-task output, tuple grader returns

Deepikachintamreddy

c7b11e5
10:08
add POST /grader endpoint for validator

Deepikachintamreddy

5e04bd7
09:29
trigger rebuild

Deepikachintamreddy

fba04fa
09:18
fix: resolve merge conflict in grader_api.py

Deepikachintamreddy

b53b055
07:31
Refactor graders to use clamping wrapper

Chintamreddy Deepika

Unverified signature
16b042a
07:30
Update grader_api.py

Chintamreddy Deepika

Unverified signature
7289a7e
07:29
Update app.py

Chintamreddy Deepika

Unverified signature
99b0783
06:42
Update env.py

Chintamreddy Deepika

Unverified signature
495f412
06:00
CRITICAL FIX: Add missing grader_api.py

Deepikachintamreddy

8ad450c

Apr 11

22:52
SIMPLIFY: Override /reset to create fresh environment directly, avoiding original endpoint call

Nanda Kumar Kondreddy

cf3a550
22:51
FIX: Properly type request parameter in /reset endpoint as fastapi.Request

Nanda Kumar Kondreddy

a91aa82
22:44
CRITICAL FIX: Replace /reset middleware with direct endpoint override - ensure base fields included

Nanda Kumar Kondreddy

f2a42f5
22:42
ENHANCE: Update reset middleware to include base Observation fields (done, reward, metadata) at top level

Nanda Kumar Kondreddy

270a32f
22:31
FIX: Add middleware to correct /reset response schema - add 'info' field, remove reward/done

Nanda Kumar Kondreddy

5bc98e9
21:57
Add comprehensive instrumentation to all graders for validator diagnostic

Nanda Kumar Kondreddy

4891046
21:05
Add detailed instrumentation to grade_task1 and normalize_reward for validator debugging

Nanda Kumar Kondreddy

31647f3
21:01
CRITICAL RUNTIME FIX: Remove tuple unpacking from environment handlers - graders now return FLOAT only, not tuples

Nanda Kumar Kondreddy

b8c69e2
20:54
CRITICAL FIX: Graders now return FLOAT ONLY (not tuples) for validator contract compliance - extracted float from raw grader tuples

Nanda Kumar Kondreddy

c256e38
20:34
CRITICAL FIX: Wrappers now return full tuples (reward, error, bugs) instead of just floats - fixes env.py unpacking

Nanda Kumar Kondreddy

cf1808d
20:30
Add comprehensive runtime grader validation endpoint (/validate-graders) to audit actual grader outputs under validator-style conditions

Nanda Kumar Kondreddy

5e5a089
20:10
CRITICAL WIRING FIX: Import wrapped graders from grader_api, not raw graders - ensures normalization is used

Nanda Kumar Kondreddy

1566ba9
19:53
FINAL FIX: Centralized universal reward normalization - GUARANTEES all scores in (0.01, 0.99)

Nanda Kumar Kondreddy

6d73738
19:30
CRITICAL BUG FIX: Clamp reward on ALL early return paths - prevents returning 0.0

Nanda Kumar Kondreddy

6138889
19:17
Enhanced /diagnostics: validate manifest import paths with importlib (mirrors validator behavior)

Nanda Kumar Kondreddy

c87f77e
19:12
Add startup and diagnostics logging to debug task loading

Nanda Kumar Kondreddy

9331aaf
18:57
Simplify metadata task schema to minimal fields (id, has_grader) for validator compatibility

Nanda Kumar Kondreddy

491de7d
18:52
PROPER FIX: Remove default /metadata route before registering custom override with tasks

Nanda Kumar Kondreddy

3a48cf2
18:43
CRITICAL FIX: Add /metadata endpoint with task enumeration for validator discovery

Nanda Kumar Kondreddy

a2f8fcd
18:30
Add validator invocation logging to all grader wrappers

Nanda Kumar Kondreddy

e924908
18:14
Add debug endpoint to verify deployed grader fix

Nanda Kumar Kondreddy

afad682
17:42
Fix: Enforce strict open interval (0.01, 0.99) for all grader rewards - validator compliance

Nanda Kumar Kondreddy

8e7ae76
17:25
Phase 2: Benchmark Quality Upgrades - Progressive graders, realistic multi-bug tasks, ground_truth exposure

Nanda Kumar Kondreddy

b43aff5

Commit History

Apr 12

fix: eliminate all 0.0 reward values across all files

fix: progress_ratio default 0.01

fix: total_reward init 0.01

fix: remove function aliases from grader_api, classes only, NaN protection

fix: grader classes with grade() method per Meta feedback

fix: model defaults 0.01 not 0.0

fix: initial previous_reward 0.01 not 0.0

fix: reset reward 0.01 not 0.0

fix: grader handles all call patterns, reset reward 0.01

fix: class-based graders with grade() method per Meta instructions

fix: bulletproof graders - never throw, always return float in range

fix: grader_api returns float only, not tuple

fix: correct step format, per-task output, tuple grader returns

add POST /grader endpoint for validator

trigger rebuild

fix: resolve merge conflict in grader_api.py

Refactor graders to use clamping wrapper

Update grader_api.py

Update app.py

Update env.py

CRITICAL FIX: Add missing grader_api.py

Apr 11

SIMPLIFY: Override /reset to create fresh environment directly, avoiding original endpoint call

FIX: Properly type request parameter in /reset endpoint as fastapi.Request

CRITICAL FIX: Replace /reset middleware with direct endpoint override - ensure base fields included

ENHANCE: Update reset middleware to include base Observation fields (done, reward, metadata) at top level

FIX: Add middleware to correct /reset response schema - add 'info' field, remove reward/done

Add comprehensive instrumentation to all graders for validator diagnostic

Add detailed instrumentation to grade_task1 and normalize_reward for validator debugging

CRITICAL RUNTIME FIX: Remove tuple unpacking from environment handlers - graders now return FLOAT only, not tuples

CRITICAL FIX: Graders now return FLOAT ONLY (not tuples) for validator contract compliance - extracted float from raw grader tuples

CRITICAL FIX: Wrappers now return full tuples (reward, error, bugs) instead of just floats - fixes env.py unpacking

Add comprehensive runtime grader validation endpoint (/validate-graders) to audit actual grader outputs under validator-style conditions

CRITICAL WIRING FIX: Import wrapped graders from grader_api, not raw graders - ensures normalization is used

FINAL FIX: Centralized universal reward normalization - GUARANTEES all scores in (0.01, 0.99)

CRITICAL BUG FIX: Clamp reward on ALL early return paths - prevents returning 0.0

Enhanced /diagnostics: validate manifest import paths with importlib (mirrors validator behavior)

Add startup and diagnostics logging to debug task loading

Simplify metadata task schema to minimal fields (id, has_grader) for validator compatibility

PROPER FIX: Remove default /metadata route before registering custom override with tasks

CRITICAL FIX: Add /metadata endpoint with task enumeration for validator discovery

Add validator invocation logging to all grader wrappers

Add debug endpoint to verify deployed grader fix

Fix: Enforce strict open interval (0.01, 0.99) for all grader rewards - validator compliance

Phase 2: Benchmark Quality Upgrades - Progressive graders, realistic multi-bug tasks, ground_truth exposure