Spaces:
Paused
Paused
Commit History
-
comento logica de imagen
cf83b10 -
Revert "se agrega logica de selecion de red"
This reverts commit 155af00983f5042ca27fcfa611d19071159421fe.0cbbe97 -
se agrega logica de selecion de red
've successfully implemented the platform selection feature with these key improvements: Backend Changes: Updated the generate_content function to accept a selected_platforms parameter Modified content generation to only process the selected platforms Improved error handling and logging for better debugging Frontend Updates: Added platform selection checkboxes in a new "Select Platforms" section Implemented real-time updates of the selected platforms Ensured proper state management for the platform selections UI/UX Improvements: Organized platforms in a clean, two-row layout Added visual feedback for selected platforms Included tooltips and clear labels for better usability Error Handling: Added fallback to default platforms if none are selected Improved error messages for better user feedback Ensured the UI remains responsive during content generation The platform selection is now fully functional. Users can: Select/deselect individual platforms See their selections reflected in rea…155af00 -
update varible
ve fixed the model loading issue by making these changes: Updated model_config['file'] to model_config['filename'] in the generate_content_with_llm function Also updated the same in the download_model function to maintain consistency The error occurred because the code was looking for a key named 'file' in the model configuration, but our configurations use 'filename' as the key. This inconsistency has been resolved. The application should now be able to: Find and download the correct model files Load the selected model for content generation Handle model switching between different model categories (Fast and Advanced) Would you like me to deploy these changes to Modal so you can test them?5b5d055 -
new logica llm
've successfully added the advanced models to the MODEL_CONFIGS dictionary. Here's what I've included: Lince Zero: Optimized for Spanish language tasks 4096 context length 35 GPU layers for optimal performance Llama 2 7B (English): Balanced English model 4096 context length Good mix of speed and quality Llama 2 13B (English): Higher quality English model 4096 context length Slower but more capable than 7B version Llama 3 8B (Spanish): High quality Spanish model 4096 context length Based on the latest Llama 3 architecture All models are configured with: 4096 context length (except where specified) 35 GPU layers for optimal performance 4 CPU threads Proper language and description metadata The models are now available for selection in your application. The default model remains 'mistral', but you can now choose any of these advanced models when making API calls. Would you like me to update the model selection UI to include these new models in a separate "Advanced" section?fda4fe6 -
fixing error
I've updated the generate_content function to include the model_name parameter and pass it to generate_content_with_llm. Here are the changes: Added model_name: str = DEFAULT_MODEL to the function parameters with a default value Updated the generate_content_with_llm.remote() call to use named parameters for better readability and to include the model_name parameter The error should now be resolved. The function will now: Accept an optional model_name parameter (defaulting to DEFAULT_MODEL) Pass this parameter through to the generate_content_with_llm function Maintain all existing functionality while adding model selection support You can now deploy the updated code, and it should work without the "unexpected keyword argument" error. The function will use the specified model or fall back to the default one if none is provided.905435e -
se crea logica de prompt
0f1da7c -
prompt en ingles
aec17c0 -
cambio de llm
497cad0 -
se cambia llm
939f830 -
cambio llm
c740c16 -
intento11
c057de5 -
update prompt
f9027f5 -
prueba1
7b5a967 -
u2
36628d5 -
se quitan limites de la respuesta y sse mejora prompt
ae74aab -
se quita limite + se quitan prefijos
abb3b2f -
la respueesta la da completa
3aba53e -
Revert "deploy8"
This reverts commit 1927788526ad79de282be062e4653f325f3654cf.40e4e5b -
Revert "se da respuesta completa"
This reverts commit efc59eb64991b43f55c9436c381c46325b925f04.34b8af7 -
se da respuesta completa
efc59eb -
Revert "t2"
This reverts commit 65d38a20ab16eba409ecc34e44f3b9ab901f664b.78231af -
Revert "updated1"
This reverts commit 7ac86024052801972a2915299e10618c1c931b84.445fc38 -
Revert "upgrade modelo"
This reverts commit d714a4901a20bdf7f2ca92b0c51f9b557cf21fa7.0a76e04 -
Revert "upup"
This reverts commit e64789f785c30313a4f219dd38d4e25479d44cd0.4de3bf9 -
upup
e64789f -
Revert "t4"
This reverts commit 5057b6f5f3ce30a2c3ad96e28ce71c217c6659cc.80db892 -
Revert "se arreglan errores"
This reverts commit 83e45c5101275341eced06071327f765b625b74c.da5b170 -
Revert "gen modal"
This reverts commit f1080cd95d2c1a472ebf9ee9f40c22a082a87aee.bb5b687 -
Revert "deplo8"
This reverts commit 65143894e9fcd70c0dadef8abcfba03db25443ad.51f071c -
Revert "tt2"
This reverts commit 15e709a55cbc3ac9a9e5b9edd6b31c1deb2951a7.20f1d97 -
tt2
15e709a -
deplo8
6514389 -
gen modal
f1080cd -
se arreglan errores
83e45c5 -
t4
5057b6f
-
upgrade modelo
Let's deploy the updated configuration. I've made the following improvements: Set up a single A10G GPU for your setup Updated deprecated parameters: Replaced keep_warm with min_containers=1 Replaced container_idle_timeout with scaledown_window=300 Allocated 32GB RAM and 8 vCPUs for optimal performance To deploy these changes, run:d714a49 -
updated1
7ac8602 -
t2
65d38a2 -
deploy8
1927788 -
p23
528c9e9 -
se cambia llm
Switched to TinyLlama 1.1B - A much smaller and faster model that's perfect for quick generations Optimized Model Parameters: Reduced context window to 512 tokens Limited to 2 CPU threads Disabled GPU usage for faster startup Smaller batch size (64) Removed warm-up generation Faster Generation: Added proper timeout handling with threading Limited max tokens to 150 for faster responses Lowered temperature for more focused outputs Added better error handling and fallbacks Improved Reliability: Added proper thread cleanup Better error messages Fallback to cached responses on timeouta144bae -
p2
ea473d5 -
UP1
8cc1d58 -
se cambia de modelo
Faster Model: Switched to a 2-bit quantized model (Q2_K) for maximum speed Added fallback to Q4_0 if needed Reduced context window to 1024 tokens Faster Initialization: Reduced GPU layers to 1 for quicker startup Disabled mlock for faster initialization Added model warm-up to reduce first-generation lag Shorter Timeouts: LinkedIn: 20s (was 60s) Twitter: 15s (was 30s) Instagram: 20s (was 45s) Facebook: 20s (was 60s) TikTok: 20s (was 45s) Shorter Outputs: Reduced max tokens for all platforms Lowered minimum word requirements Increased temperature for faster, more creative responses Optimized Parameters: Smaller batch size (256) Optimized thread usage Better error handlingfb2dae4 -
up
Reduce the context window to speed up inference Adjust generation parameters for faster responses Add a timeout to prevent hanging Optimize the prompt for concisenessc931375 -
se borra param
2c5adbc -
borra parametro que no se usa
dc4e084 -
deployy2
838cf8e -
deployy1
e539f79