0.0.188

We’ve got a fresh batch of improvements rolling out with the new release: 0.0.188 — dropping tomorrow at 09:00 UTC.

🔍 Twitter API Upgrades
• Added view count + replies to the protocol — you now get the full X post object
• Improved view + quote scoring logic, forcing miners to return better, more complete data
• Query parser now handles filters like bittensor min_faves:100 — these get scored properly and give you more powerful Twitter search control

🤖 Advanced AI Search Enhancements
• Custom scoring system messages: You can now define your own scoring rules per request. Want to prioritize a certain topic, tone, or context? Just set the prompt — Twitter and Web scorers will adapt. No more relying solely on default validator logic.
• Chat history scoring improvements: We’ve added smart penalty logic based on message history. This helps provide better answers with full context across conversations — improving search results and user experience in long chats.

⚙️ Other Fixes
• SN64 scoring has been fully migrated to an async HTTP library — better performance and faster response times

0.0.187

we've made significant updates to our subnet, making it more expansive and flexible for our users. These updates include important changes, bug fixes, improvements, and new features.

Here's a detailed look at what's new:

🛠️ Integration with SN64 for Validators: We're enabling validators to use SN64 models for scoring, which will reduce costs and increase the use of open-source models. The models we're supporting include:

Qwen/Qwen2.5-Coder-32B-Instruct
Mistral-Small-24B-Instruct-2501
DeepSeek-R1-Distill-Qwen-32B
We've tested multiple models and selected these based on their accuracy and performance compared to GPT-4o.

📝 User System Message: Based on user feedback, we've added a feature that allows users to generate summaries and define rules themselves.

⚙️ Scoring Mechanism Improvements: We've improved, updated, and fixed bugs in the scoring mechanism across all directions.

🔍 Reduction in Synthetic Queries: We've reduced the number of synthetic queries and focused scoring on organic queries.

🔗 New Features for Miners: Miners will now return links with scores, and validators should validate them.

📈 Endpoint Scoring Updates: We've fixed and updated the scoring for new endpoints.

🔧 Web Search API: We have a Basic Web API and have added scoring for validators. Miners must return responses on the web API.

🔄 Bittensor Update: We've updated to the latest version, 9.0.3.

🗄️ Redis Storage: Miner results and scoring history will now be stored in Redis for validators.

⚖️ Weight Logic Update: We've updated the logic for setting weights.

⚖️ Search Tweets By URL and ID: Improve scoring scoring for each tweets.

0.0.182

Merge pull request #271 from Desearch-ai/fix/resync-metagraph

Fix: resync metagraph, add subtensor error handling

0.0.102

feat: log original relevance scores

0.0.58

fix: stream buffer issue

0.0.29

Here are some updates:
[x] We have added support for GPT-3.5 on the miner's side, which is now the default model.

It is utilized in three scenarios:

• --miner.openai_summary_model: OpenAI model used for summarizing content. Default: gpt-3.5-turbo-1106
• --miner.openai_query_model: OpenAI model used for generating queries. Default: gpt-3.5-turbo-1106
• --miner.openai_fix_query_model: OpenAI model used for fixing queries. Default: gpt-4-1106-preview

For more details, visit: https://github.com/surcyf123/smart-scrape/blob/main/docs/running_a_miner.md

You can use any OpenAI models, With the default Models you can reduce OpenAI cost 7-10 times.

[x] Updated the mockdataset class and added more synthetic data questions.
[x] Changed the synthetic data interval from 10 minutes to 30 minutes.
[x] Refactor code

0.0.25

Twitter API Improvements:

[x] The miner will not start if the TWITTER_BEARER_TOKEN, OPENAI_API_KEY, or WANDB_API_KEY have not been set.
[x] If, while running, the miner receives a 401 or 403 error from the Twitter API, it will stop working. In this case, you need to promptly address your API key issues to get your miner back on the network.
[x] If, while running, the miner receives a 400 Bad Request error from the Twitter API, it will attempt to fix the query and retrieve data again.
[x] If, while running, the miner receives 429, 502, 503, or 504 errors from the Twitter API, it will wait for 20 seconds and then make one more attempt to connect to the Twitter API.
[x] Update the retry attempt prompt and methods in the Twitter API.