Spaces:
Sleeping
Sleeping
Commit ·
c372ed2
1
Parent(s): 0e31d8c
added openenv such that validation get's passed
Browse files- Dockerfile +1 -1
- openenv.yaml +84 -0
Dockerfile
CHANGED
|
@@ -2,7 +2,7 @@ FROM python:3.11-slim
|
|
| 2 |
|
| 3 |
WORKDIR /app
|
| 4 |
|
| 5 |
-
COPY requirements.txt
|
| 6 |
RUN pip install --no-cache-dir -r requirements.txt
|
| 7 |
RUN python -c "import nltk; nltk.download('vader_lexicon', quiet=True); nltk.download('punkt_tab', quiet=True)"
|
| 8 |
|
|
|
|
| 2 |
|
| 3 |
WORKDIR /app
|
| 4 |
|
| 5 |
+
COPY requirements.txt .
|
| 6 |
RUN pip install --no-cache-dir -r requirements.txt
|
| 7 |
RUN python -c "import nltk; nltk.download('vader_lexicon', quiet=True); nltk.download('punkt_tab', quiet=True)"
|
| 8 |
|
openenv.yaml
ADDED
|
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
name: sieve
|
| 2 |
+
version: "1.0.0"
|
| 3 |
+
description: >
|
| 4 |
+
A customer support reinforcement learning environment where an agent triages,
|
| 5 |
+
classifies, drafts responses to, and prioritizes real-world support emails.
|
| 6 |
+
tags:
|
| 7 |
+
- openenv
|
| 8 |
+
- customer-support
|
| 9 |
+
- email-triage
|
| 10 |
+
- nlp
|
| 11 |
+
|
| 12 |
+
api:
|
| 13 |
+
base_url: http://localhost:7860
|
| 14 |
+
reset: POST /reset
|
| 15 |
+
step: POST /step
|
| 16 |
+
state: GET /state
|
| 17 |
+
tasks: GET /tasks
|
| 18 |
+
|
| 19 |
+
tasks:
|
| 20 |
+
- id: email_classification
|
| 21 |
+
name: Email Classification
|
| 22 |
+
difficulty: easy
|
| 23 |
+
description: >
|
| 24 |
+
Classify each incoming customer support email by category
|
| 25 |
+
(billing/technical/general/spam/account/feature_request)
|
| 26 |
+
and urgency (high/medium/low) using action_type='classify'.
|
| 27 |
+
max_steps: 15
|
| 28 |
+
score_range: [0.0, 1.0]
|
| 29 |
+
|
| 30 |
+
- id: response_drafting
|
| 31 |
+
name: Response Drafting
|
| 32 |
+
difficulty: medium
|
| 33 |
+
description: >
|
| 34 |
+
Draft professional, empathetic responses to customer support emails.
|
| 35 |
+
Each response must cover required keywords, exceed 50 characters,
|
| 36 |
+
and maintain a professional tone. Use action_type='respond'.
|
| 37 |
+
max_steps: 10
|
| 38 |
+
score_range: [0.0, 1.0]
|
| 39 |
+
|
| 40 |
+
- id: support_session
|
| 41 |
+
name: Full Support Session
|
| 42 |
+
difficulty: hard
|
| 43 |
+
description: >
|
| 44 |
+
Manage a queue of 15 mixed emails. Prioritize VIP customers first,
|
| 45 |
+
then high-urgency emails. Choose the correct action (respond/escalate/archive)
|
| 46 |
+
per email, provide category and urgency, and use email_id to select
|
| 47 |
+
which email to process each step.
|
| 48 |
+
max_steps: 40
|
| 49 |
+
score_range: [0.0, 1.0]
|
| 50 |
+
|
| 51 |
+
observation_space:
|
| 52 |
+
current_email:
|
| 53 |
+
id: string
|
| 54 |
+
subject: string
|
| 55 |
+
body: string
|
| 56 |
+
sender: string
|
| 57 |
+
sender_tier: "standard | vip"
|
| 58 |
+
received_minutes_ago: integer
|
| 59 |
+
email_queue: "array of Email objects (populated in support_session only)"
|
| 60 |
+
processed_count: integer
|
| 61 |
+
step_count: integer
|
| 62 |
+
task_id: string
|
| 63 |
+
task_description: string
|
| 64 |
+
available_actions: "array of strings"
|
| 65 |
+
context:
|
| 66 |
+
max_steps: integer
|
| 67 |
+
remaining_steps: integer
|
| 68 |
+
queue_size: integer
|
| 69 |
+
|
| 70 |
+
action_space:
|
| 71 |
+
action_type: "classify | respond | escalate | archive | skip"
|
| 72 |
+
category: "billing | technical | general | spam | account | feature_request"
|
| 73 |
+
urgency: "high | medium | low"
|
| 74 |
+
response_text: "string — required for respond"
|
| 75 |
+
escalation_reason: "string — required for escalate"
|
| 76 |
+
email_id: "string — used in support_session to select target email"
|
| 77 |
+
|
| 78 |
+
baseline:
|
| 79 |
+
agent: gpt-4o-mini
|
| 80 |
+
scores:
|
| 81 |
+
email_classification: 0.860
|
| 82 |
+
response_drafting: 0.956
|
| 83 |
+
support_session: 0.850
|
| 84 |
+
average: 0.889
|