Skip to content

Speakr is a personal, self-hosted web application designed for transcribing audio recordings

Python

fixed context length issues

Murtaza Nasirfa9c9add99ff
diff --git a/.env.example b/.env.example
index e010d738..ea0f5505 100644
--- a/.env.example
+++ b/.env.example
@@ -15,6 +15,10 @@ WHISPER_MODEL="Systran/faster-distil-whisper-large-v3"
# Set to "false" to disable new account registration
ALLOW_REGISTRATION="false"
+# Max tokens for LLM responses (optional, defaults are set in app.py)
+SUMMARY_MAX_TOKENS="8000"
+CHAT_MAX_TOKENS="5000"
+
# Models
# openai/gpt-4o-mini
# google/gemma-3-27b-it
diff --git a/.gitignore b/.gitignore
index 3d9d256c..cba066d5 100644
--- a/.gitignore
+++ b/.gitignore
@@ -12,4 +12,5 @@ __pycache__/
project_files.md
*.md
notes.md
-docker-compose.yml
\ No newline at end of file
+docker-compose.yml
+changes.txt
diff --git a/app.py b/app.py
index 9478d41c..048a5ff7 100644
--- a/app.py
+++ b/app.py
@@ -390,7 +390,7 @@ JSON Response:"""
{"role": "user", "content": prompt_text}
],
temperature=0.5, # Adjust temperature as needed
- max_tokens=300, # Adjust based on expected title+summary length
+ max_tokens=int(os.environ.get("SUMMARY_MAX_TOKENS", "3000")), # Adjust based on expected title+summary length
response_format={"type": "json_object"} # Request JSON output
)
@@ -565,7 +565,7 @@ Additional context and notes about the meeting:
model=TEXT_MODEL_NAME,
messages=messages,
temperature=0.7,
- max_tokens=1000
+ max_tokens=int(os.environ.get("CHAT_MAX_TOKENS", "2000"))
)
response_content = completion.choices[0].message.content
diff --git a/docker-compose.example.yml b/docker-compose.example.yml
index 361d9716..c7207625 100644
--- a/docker-compose.example.yml
+++ b/docker-compose.example.yml
@@ -41,6 +41,8 @@ services:
# Application settings
# Transcription and Output languages are now per-user settings, configurable on the Account page in the app.
- ALLOW_REGISTRATION=false
+ - SUMMARY_MAX_TOKENS=8000 # Optional: Max tokens for title/summary generation (default: 3000)
+ - CHAT_MAX_TOKENS=5000 # Optional: Max tokens for chat responses (default: 2000)
# Admin user creation
- ADMIN_USERNAME=admin
diff --git a/readme.md b/readme.md
index 47538274..ef6efe08 100644
--- a/readme.md
+++ b/readme.md
@@ -364,6 +364,8 @@ Configuration is primarily handled through environment variables in the `docker-
* `TRANSCRIPTION_BASE_URL`: **Required.** The base URL for your transcription API endpoint (e.g., `http://localhost:8787/v1/`).
* `TRANSCRIPTION_API_KEY`: **Required.** Your API key for the transcription endpoint. For local endpoints, this might be a specific string like "none" or "NA". Check your endpoint's documentation.
* `WHISPER_MODEL`: *Optional.* The specific model name your transcription endpoint uses/expects (e.g., `Systran/faster-distil-whisper-large-v3`). Check your endpoint's requirements.
+* `SUMMARY_MAX_TOKENS`: *Optional.* The maximum number of tokens for the title and summary generation. Defaults to `3000`.
+* `CHAT_MAX_TOKENS`: *Optional.* The maximum number of tokens for chat responses. Defaults to `2000`.
* `SECRET_KEY`: **Required.** A long, random string used by Flask for session security. The `setup.sh` script generates one if it's missing.
* `ALLOW_REGISTRATION`: *Optional.* Set to `false` to prevent new users from registering via the web UI. Defaults to `true`.