Comments (10)
Ok, I'll test it if I have time
from openchat.
We started the OpenAI API server and ran the evaluation using alpaca_eval with a context length of 4096. Most of our evaluations are around 89%, with a standard deviation of about 1%.
BTW, are you using alpaca_eval_gpt4
or alpaca_eval_gpt4_fn
? The latter would produce 5% lower results similar to what you observed.
from openchat.
We started the OpenAI API server and ran the evaluation using alpaca_eval with a context length of 4096. Most of our evaluations are around 89%, with a standard deviation of about 1%.
BTW, are you using
alpaca_eval_gpt4
oralpaca_eval_gpt4_fn
? The latter would produce 5% lower results similar to what you observed.
Many thanks for your kind response! I doulbe check my evaluation setting; I confirm I am using alpaca_eval_gpt4
. This is very strange, since I do not run any inference on my own and just use the model_output.json
you released on alpaca_eval.
Is this related to gpt-4 secretly update their api? Have you tested the results recently?
from openchat.
Chances are, GPT4 is constantly being updated. The previous results were all around 89%. I haven't tested it recently.
BTW, have you checked the number of rated responses? Sometimes assessments are incomplete.
from openchat.
Chances are, GPT4 is constantly being updated. The previous results were all around 89%. I haven't tested it recently.
BTW, have you checked the number of rated responses? Sometimes assessments are incomplete.
I understand your points. I double check the rated responses, the total number of rated responses are 804, with 3 of them are rated as tie.
from openchat.
@REIGN12 I've tested today using the output file, got 0.01% higher results
openchat-v3.1-13b 89.50 1.08 805 1484.00
from openchat.
OPENAI_API_KEYS=sk-XXX PYTHONPATH="$PYTHONPATH:$(pwd)/src" python -m alpaca_eval.main evaluate --model_outputs results/openchat-v3.1-13b/model_outputs.json --annotators_config alpaca_eval_gpt4
from openchat.
@REIGN12 I've tested today using the output file, got 0.01% higher results
openchat-v3.1-13b 89.50 1.08 805 1484.00
Many thanks for your reply! So here the results are tested with gpt-4-0314? I am going to double check my gpt-4 api version.
from openchat.
@REIGN12 I've tested today using the output file, got 0.01% higher results
openchat-v3.1-13b 89.50 1.08 805 1484.00
Many thanks for your reply! So here the results are tested with gpt-4-0314? I am going to double check my gpt-4 api version.
I double check my api version, I am using gpt-4-0314.
from openchat.
This is gpt-4
(latest), e.g. the alpaca_eval_gpt4
evaluator. I evaluated with the latest alpaca_eval and following commands:
OPENAI_API_KEYS=sk-XXX PYTHONPATH="$PYTHONPATH:$(pwd)/src" python -m alpaca_eval.main evaluate --model_outputs results/openchat-v3.1-13b/model_outputs.json --annotators_config alpaca_eval_gpt4
from openchat.
Related Issues (20)
- What is a minimum hardware requirements for setup as a sever ?
- Shall we add a Contrib Guide .md (linked and separate from README)?
- in llama.cpp load gguf has some issue in it
- How was openchat/openchat-3.5-0106-gemma created? HOT 2
- Error running new model openchat-3.5-0106-gemma on RTX 4090 24GB machine, works with older mistral based model HOT 2
- llama_model_load: error loading model: create_tensor: tensor 'output.weight' not found HOT 1
- Was the chat template applied in ochat/config/conversation_template.py?
- openchat.team is down for 3 days HOT 1
- Error when using openchat/openchat-3.5-0106-gemma in text-generation-inference
- WARNING: Error in configuration: macro '\frac' failed its substitution!
- weighted_token_accuracy
- about data
- For ๐ปOnline Demo It has been broken out. HOT 2
- Single GPU vs multiple GPU (tensor parallel) suggestion for API Server HOT 3
- repeat the output content until the maximum output length is set HOT 3
- A question about the prefix of only SFT HOT 1
- opchatdataset.estimate_num_batches returns 0 at the beginning of training, and training-stuck problem HOT 1
- use instructor in openchat HOT 1
- huggingface down: openchat can't be started HOT 1
- About performance of LLama (llama2, llama3) model HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
๐ 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. ๐๐๐
-
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 โค๏ธ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from openchat.