Update README.md

README.md

@@ -55,13 +55,8 @@ generation_config = GenerationConfig(
     max_new_tokens=1536
 )
 
-instruction = """Using ABC notation, recreate the given text as a musical score.
-Meter C
-Notes The parts are commonly interchanged.
-Transcription 1997 by John Chambers
-Key D
-Note Length 1/8
-Rhythm reel
+instruction = """Develop a musical piece using the given chord progression.
+'Dm', 'C', 'Dm', 'Dm', 'C', 'Dm', 'C', 'Dm'
 """
 
 prompt = prompt_template.safe_substitute({"inst": instruction})
@@ -87,10 +82,11 @@ dump_wav('cm_music_piece.wav', audio, sample_rate=44100, use_int16=True)
 ```
 
 ## Limitations
-- For the music tasks, the model currently only supports strict format and close-ended instructions.
-- The model suffers from hallucinations, and shouldn't be used for music education.
-- A large proportion of the training data is in the style of Irish music.
-- The MusicThoeryBench results reported in the paper are obtained with perplexity mode. 
+- For the music tasks, the model currently only supports strict format and close-ended instructions. If we have more funding, we plan to create a more diverse multi-turn music instruction chat data for better generalization.
+- The model suffers from hallucinations, and shouldn't be used for music education. It could be improved by feeding more music textbooks, blogs, etc. And RLHF may help, too.
+- A large portion of the training data is in the style of Irish music. If possible, the community should develop a converter between performance midi and ABC scores.
+- The MusicThoeryBench results reported in the paper are obtained with perplexity mode. Direct generation may result in a worse performance.
+- We observe that using current version of training data, ChatMusician may lose its in-context-learning and chain-of-thoughts ability. The community should work on improving the music data quality. 
 
 ## Example Prompts