The link of the source code is here: https://github.com/google-deepmind/alphafold3/releases
\nI have tried so many times, and it seems like that it need python 3.11, kaggle's python version is 3.10, can we overcome this to run AlphaFold 3?
", "messages": [ { "id": "3157584", "postDate": "03/23/2025 15:54:42", "content": "The link of the source code is here: https://github.com/google-deepmind/alphafold3/releases
\nI have tried so many times, and it seems like that it need python 3.11, kaggle's python version is 3.10, can we overcome this to run AlphaFold 3?
", "rawMarkdown": "The link of the source code is here: https://github.com/google-deepmind/alphafold3/releases\n\nI have tried so many times, and it seems like that it need python 3.11, kaggle's python version is 3.10, can we overcome this to run AlphaFold 3?", "votes": null }, { "id": "3157866", "postDate": "03/24/2025 00:58:42", "content": "You can try Protenix: https://github.com/bytedance/Protenix
", "rawMarkdown": "You can try Protenix: https://github.com/bytedance/Protenix", "votes": null }, { "id": "3158296", "postDate": "03/24/2025 12:15:56", "content": "Thanks! Can try it :)
", "rawMarkdown": "Thanks! Can try it :)", "votes": null }, { "id": "3158852", "postDate": "03/25/2025 01:41:22", "content": "Hi! May I ask if you know how to download the model for prediction as it is not in the github?
", "rawMarkdown": "Hi! May I ask if you know how to download the model for prediction as it is not in the github?", "votes": null }, { "id": "3158853", "postDate": "03/25/2025 01:44:01", "content": "My error is like this:
\nš§¬ Predicting with Protenix: 92%|āāāāāāāāāā| 11/12 [00:41<00:03, 3.76s/it]
\nā Error in R1108: Error(s) in loading state_dict for Protenix:
\n size mismatch for confidence_head.linear_no_bias_d.weight: copying a param with shape torch.Size([128, 15]) from checkpoint, the shape in current model is torch.Size([128, 39]).
\ntrain scheduler 16.0
\ninference scheduler 16.0
\nDiffusion Module has 16.0
You can fork the github repository and check the file inference_demo.sh
export LAYERNORM_TYPE=fast_layernorm\nexport USE_DEEPSPEED_EVO_ATTENTION=true\n\nN_sample=\nN_step=\nN_cycle=\nseed=\nuse_deepspeed_evo_attention=true\ninput_json_path=\ndump_dir=\n\npython3 runner/inference.py \\\n--seeds ${seed} \\\n--dump_dir ${dump_dir} \\\n--input_json_path ${input_json_path} \\\n--model.N_cycle ${N_cycle} \\\n--sample_diffusion.N_sample ${N_sample} \\\n--sample_diffusion.N_step ${N_step}\nYou used pip then protenix predict --input examples/example.json --out_dir ./output --seeds 101, I guess?
Hi! May I ask if you can do the inference only using the coding from github or not? Do you need to download extra model? (I download one which is 1.5GB but not match)
", "rawMarkdown": "Hi! May I ask if you can do the inference only using the coding from github or not? Do you need to download extra model? (I download one which is 1.5GB but not match)", "votes": null }, { "id": "3159838", "postDate": "03/26/2025 01:14:51", "content": "I trained my own model, not only github itself.
", "rawMarkdown": "I trained my own model, not only github itself.", "votes": null } ], "comments": [ { "id": 3157866, "author_name": "sweetyheehee", "author_url": "", "post_date": "03/24/2025 00:58:42", "content": "You can try Protenix: https://github.com/bytedance/Protenix
", "votes": null, "replies": [ { "id": 3158296, "author_name": "larrylin666", "author_url": "", "post_date": "03/24/2025 12:15:56", "content": "Thanks! Can try it :)
", "votes": null, "replies": [] }, { "id": 3158852, "author_name": "larrylin666", "author_url": "", "post_date": "03/25/2025 01:41:22", "content": "Hi! May I ask if you know how to download the model for prediction as it is not in the github?
", "votes": null, "replies": [] }, { "id": 3158853, "author_name": "larrylin666", "author_url": "", "post_date": "03/25/2025 01:44:01", "content": "My error is like this:
\nš§¬ Predicting with Protenix: 92%|āāāāāāāāāā| 11/12 [00:41<00:03, 3.76s/it]
\nā Error in R1108: Error(s) in loading state_dict for Protenix:
\n size mismatch for confidence_head.linear_no_bias_d.weight: copying a param with shape torch.Size([128, 15]) from checkpoint, the shape in current model is torch.Size([128, 39]).
\ntrain scheduler 16.0
\ninference scheduler 16.0
\nDiffusion Module has 16.0
You can fork the github repository and check the file inference_demo.sh
export LAYERNORM_TYPE=fast_layernorm\nexport USE_DEEPSPEED_EVO_ATTENTION=true\n\nN_sample=\nN_step=\nN_cycle=\nseed=\nuse_deepspeed_evo_attention=true\ninput_json_path=\ndump_dir=\n\npython3 runner/inference.py \\\n--seeds ${seed} \\\n--dump_dir ${dump_dir} \\\n--input_json_path ${input_json_path} \\\n--model.N_cycle ${N_cycle} \\\n--sample_diffusion.N_sample ${N_sample} \\\n--sample_diffusion.N_step ${N_step}\nYou used pip then protenix predict --input examples/example.json --out_dir ./output --seeds 101, I guess?
Hi! May I ask if you can do the inference only using the coding from github or not? Do you need to download extra model? (I download one which is 1.5GB but not match)
", "votes": null, "replies": [ { "id": 3159838, "author_name": "sweetyheehee", "author_url": "", "post_date": "03/26/2025 01:14:51", "content": "I trained my own model, not only github itself.
", "votes": null, "replies": [] } ] } ] } ] } ] } ], "raw_markdown_by_id": { "3157584": "The link of the source code is here: https://github.com/google-deepmind/alphafold3/releases\n\nI have tried so many times, and it seems like that it need python 3.11, kaggle's python version is 3.10, can we overcome this to run AlphaFold 3?", "3157866": "You can try Protenix: https://github.com/bytedance/Protenix", "3158296": "Thanks! Can try it :)", "3158852": "Hi! May I ask if you know how to download the model for prediction as it is not in the github?", "3158853": "My error is like this:\nš§¬ Predicting with Protenix: 92%|āāāāāāāāāā| 11/12 [00:41<00:03, 3.76s/it]\nā Error in R1108: Error(s) in loading state_dict for Protenix:\n\tsize mismatch for confidence_head.linear_no_bias_d.weight: copying a param with shape torch.Size([128, 15]) from checkpoint, the shape in current model is torch.Size([128, 39]).\ntrain scheduler 16.0\ninference scheduler 16.0\nDiffusion Module has 16.0", "3158858": "You can fork the github repository and check the file `inference_demo.sh`\n\n```python\nexport LAYERNORM_TYPE=fast_layernorm\nexport USE_DEEPSPEED_EVO_ATTENTION=true\n\nN_sample=5\nN_step=200\nN_cycle=10\nseed=101\nuse_deepspeed_evo_attention=true\ninput_json_path=\"./examples/example.json\"\ndump_dir=\"./output\"\n\npython3 runner/inference.py \\\n--seeds ${seed} \\\n--dump_dir ${dump_dir} \\\n--input_json_path ${input_json_path} \\\n--model.N_cycle ${N_cycle} \\\n--sample_diffusion.N_sample ${N_sample} \\\n--sample_diffusion.N_step ${N_step}\n```", "3158859": "You used pip then `protenix predict --input examples/example.json --out_dir ./output --seeds 101`, I guess?", "3159781": "Hi! May I ask if you can do the inference only using the coding from github or not? Do you need to download extra model? (I download one which is 1.5GB but not match)", "3159838": "I trained my own model, not only github itself." }, "source": "meta" }