Comments (3)
If you are interested in a research project, you might have all of the relevant tools necessary already at your disposal. There is a recent paper about using a LLM to interpret internal model parameters. If you can use an already trained LLM and give it access to the model parameters and the reviews (which are a freely available collection of game reviews from the Go Teaching Ladder) then you might be able to train the LLM to comment games and explain moves as a human would.
However, this is a non-trivial task and a non-trivial ask. I would be interested if you make anything out of it, using either Gemma or OPT or Llama2 or whichever.
from katago.
Probably not. General-purpose LLMs right now aren't going to be very good at Go, and will have almost no training data for interpreting the stats of a tree of nodes from MCTS in Go.
from katago.
Probably not. General-purpose LLMs right now aren't going to be very good at Go, and will have almost no training data for interpreting the stats of a tree of nodes from MCTS in Go.
Is there any other possibility then? We know that Go AI has achieved a certain 'god-like' level through extensive self-play. By analogy, could we apply a fine-tuning approach similar to the one used for large language models (LLMs) to Katago? In this case, the objective would be to make Katago understand human game records (even those of specific players) based on its existing model parameters, with the goal of identifying the positions where humans are most prone to making mistakes.
from katago.
Related Issues (20)
- Please tell me how to load 40b weights with the latest katago engine 1.14 and load 40b weights with engine 1.13. Is there any difference between chess strength or chess?
- Possible bug: GPU may be triggered by accident while using Eigen backend HOT 3
- does katago cuda version run with zluda?
- openclUseFP16 testing not works well on AMD-Integrated GPU (Using OpenCL Device 0: gfx90c (Advanced Micro Devices, Inc.) ) HOT 2
- Interface for a custom algorithm to play as a bot HOT 8
- Katago model preference for Go beginner HOT 16
- error CL_OUT_OF_RESOURCES when launching ./katago benchmark
- Can't compile for OpenCL 1.2 HOT 3
- FATAL ERROR: Failed test assert: buf.result->policyProbs[buf.result->getPos(Location::ofString("E16",board),board)] >= 0.95 HOT 2
- ERROR: task loop loop thread failed: Got nonfinite for policy sum HOT 11
- Scoring bug HOT 2
- Unable to compile - error: invalid conversion from ‘float**’ to ‘void**’
- Pytorch load model from tensorflow .ckpt and play: Mismatch and Performance Issue with b18c384nbt model_pytorch and Checkpoint File HOT 2
- How is distributed training implemented in KataGo? HOT 1
- A question in the "match" module HOT 4
- Loading the latest 28b engine failed with 8 4090 cards HOT 2
- A question about moves not being buffed by LCB HOT 2
- match: Wrong useNHWC/FP16 settings are given to bot if there are bot dedups HOT 2
- difference in winrate and scorelead between MoveInfos and RootInfo HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from katago.