about {path-to-diffusion-lm}
lcy5058 opened this issue · 8 comments
When finally generating controllable text, python scripts/infill.py --model_path {path-to-diffusion-lm} command, which path does "path-to-diffusion-lm" refer to, the first step model path, or the last but one model generated from the classifier?
"improved diffusion/diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e" or "Classifier_models/e2e-tgt-tree_e=6_b=10_m=bert base uncased_wikitext-103-raw-v1_101_wp_None"
Hello, May I ask what path did you use in the decode Diffusion-LM part?
Hello, May I ask what path did you use in the decode Diffusion-LM part?
improved diffusion/diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e
Hello, May I ask what path did you use in the decode Diffusion-LM part?
improved diffusion/diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e
I am using the same path but the printed output list is empty. Have you met the same problem?
Hello, May I ask what path did you use in the decode Diffusion-LM part?
improved diffusion/diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e
I am using the same path but the printed output list is empty. Have you met the same problem?
The path is "diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e"
there is nothing special about batch_decode. It's a helper function that calls the right command. If it doesn't work, my recommendation is to run the command directly:
python scripts/text_sample.py --model_path diffusion_models/diff_roc_pad_rand128_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd101_xstart_e2e/ema_0.9999_100000.pt --batch_size 50 --num_samples 50 --top_p 1.0 --out_dir genout
if you want to figure out why it doesn't work, I guess (since you return empty list), there might be something about this try/except
to answer another question about which model path to use for controllable generation. you use the Diffusion-LM model path (the first one). To pass in the control classifier path, you might want to modify in the code (sorry for the bad coding practice, I will clean this up in the next version of the code release), e.g., you can search from the string "from_pretrained" to narrow the search space. In the case of Tree, it should be line 316.
Hello! Thank you very much for sharing. I would like to ask which file are you using in the path:
"diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e"
that you mentioned?
There are five files here: log, progress, random_emb.torch, training_args.json, vocab.json.
I only enter the path "diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e" ,
the following error will occur:
IsADirectoryError: [Errno 21] Is a directory: 'diffusion_models/diff_e2e-tgt_block_rand16_transformer_lr0.0001_0.0_2000_sqrt_Lsimple_h128_s2_d0.1_sd102_xstart_e2e'
I would appreciate it if you could take some time to answer my questions. Thanks.
I had the same problem as above.
I would appreciate it if you could take some time to answer my questions. Thanks.