TT-Bio

████████╗████████╗        ██████╗  ██╗  ██████╗
╚══██╔══╝╚══██╔══╝        ██╔══██╗ ██║ ██╔═══██╗
   ██║      ██║    █████╗ ██████╔╝ ██║ ██║   ██║
   ██║      ██║    ╚════╝ ██╔══██╗ ██║ ██║   ██║
   ██║      ██║           ██████╔╝ ██║ ╚██████╔╝
   ╚═╝      ╚═╝           ╚═════╝  ╚═╝  ╚═════╝

[!IMPORTANT] TT-Boltz is now TT-Bio

TT-Bio runs Boltz-2, ESMFold2, Protenix-v2, and OpenDDE structure prediction, BoltzGen binder design, ProteinMPNN sequence design, and ESMC protein embeddings, and SaProt structure-aware protein embeddings on Tenstorrent Blackhole and Wormhole, supporting single-card and multi-card configurations (e.g. QuietBox with 4 cards or Galaxy server with 32 cards). Multiple machines can also be combined into a single prediction run.

Installation

Create a Python virtual environment with Python 3.10 or 3.12, install, then install the matching Tenstorrent system dependencies.

python3.10 -m venv env
source env/bin/activate
pip install tt-bio
tt-bio install-deps

tt-bio install-deps installs the Tenstorrent system dependencies that match this release. It may ask for your sudo password.

From GitHub / source

Pin to a tagged release, track nightly main (may be untested), or work from an editable clone:

pip install "tt-bio @ git+https://github.com/moritztng/tt-bio.git@v0.3.1"   # pinned release, see Releases for the latest
pip install "tt-bio @ git+https://github.com/moritztng/tt-bio.git@main"     # nightly
# or
git clone https://github.com/moritztng/tt-bio.git
cd tt-bio
pip install -e .
tt-bio install-deps

Optional: Build TT-Metal / TT-NN from Source

If you need to build from source, follow the Tenstorrent Installation Guide.

Verify Installation

tt-bio --help
tt-bio predict --help
tt-bio msa --help

Basic Usage

Structure Prediction

tt-bio predict examples/prot.yaml --model boltz2 --override

Every command names its model with --model:

• boltz2: folds complexes of proteins, DNA, RNA, and ligands and predicts binding affinity. MSA-dependent (uses an MSA by default).
• esmfold2 / esmfold2-fast: fold a single protein sequence on-device, no MSA required (esmfold2-fast is the lighter, faster checkpoint).
• protenix-v2: folds complexes of proteins, RNA, DNA, and ligands (an AlphaFold3-family model, the Protenix reproduction); MSA-dependent for proteins (uses an MSA by default), and also emits a PAE/PDE matrix with --write_pae.
• saprot: structure-aware protein embeddings — an ESM-2 encoder over a fused amino-acid + Foldseek-3Di vocabulary (446 tokens). Needs a structure for the 3Di structural tokens (--structure); runs sequence-only without it. Use for variant-effect / mutation-fitness scoring and function prediction.
• opendde / opendde-abag: antibody-antigen co-folding built on the Protenix-v2 stack plus a structural-token expander; opendde-abag selects the antibody-antigen checkpoint. Protein-only for now; proteins are MSA-dependent (uses an MSA by default, like Protenix-v2).

tt-bio predict examples/prot.fasta --model esmfold2-fast --fast
tt-bio predict examples/prot.yaml --model protenix-v2   # MSA on by default; NA/ligand chains are single-sequence
tt-bio predict examples/9dsg_abag.yaml --model opendde-abag   # antibody-antigen co-fold, MSA on by default

All structure models support the sampling, output-format, and scheduling options. MSA, affinity, constraint, and auxiliary-output options apply only where listed below. Each model downloads its weights automatically on first use.

Boltz-2, Protenix-v2, and OpenDDE are MSA-dependent and use an MSA by default, a local ColabFold DB (~/.boltz/msa_db) if one is set up (see Offline MSA), otherwise the online ColabFold server. Sending sequences to the online server (api.colabfold.com) leaves your machine; a one-line notice is printed when that fallback is used. Pass --msa_db_path for a private offline database, or --single_sequence to deliberately fold without an MSA (lower accuracy; for batch-screening orphan sequences). OpenDDE multi-chain predictions still request paired MSAs from --msa_server_url; use --single_sequence to prevent all network MSA requests. ESMFold2 is single-sequence.

--fast makes some operations use a lower-precision numeric format that runs faster. Accuracy is typically very close.

OpenDDE-abag matches the upstream checkpoint on the standard 1AHW antibody-antigen target. Both implementations perform poorly on 9DSG.

predict accepts either a single YAML/FASTA file or a directory containing many input files.

A live display shows the progress of each target. Prediction uses up to one card per pending target, labelled in the display (quietbox:tt0, quietbox:tt1, ...). Models load once per active card and stay resident:

tt-bio predict proteins/ --model boltz2 --out_dir results --fast

Pass --devices 0,1,2,3 to pick or limit the available cards. A single target remains a single-card fold; additional cards increase throughput only when multiple targets are queued.

If you have additional machines with Tenstorrent cards, you can add them to a single run; see Optional: Multi-Machine Prediction.

Protein Embeddings (ESMC)

Turn protein sequences into ESMC language-model embeddings on-device (no folding, no MSA). DATA is a FASTA file, a directory of them, a YAML {id: sequence} mapping, or a bare sequence string:

tt-bio embed proteins.fasta --model esmc-600m --out_dir embeddings
tt-bio embed "MQIFVKTLTGKTITLEV..." --model esmc-600m   # one-off sequence

--model selects the ESMC variant (esmc-300m, esmc-600m, esmc-6b). For each sequence you get its per-residue embeddings ([length, d_model] float32, one row per amino acid, row order == input order) and a pooled whole-sequence vector ([d_model] float32, --pool mean/max/cls). --out_dir (default ./embeddings) gets:

• <id>.npz per sequence: per_residue, pooled (+ logits with --logits); --format npz, default
• embeddings.parquet: pooled vectors, one row per sequence; --format parquet
• manifest.json: model/pool/shapes/dtype and which file holds each sequence

Add --logits for the per-residue amino-acid predictions (300M/600M only), and --fast for the lower-precision weight path. Weights download automatically on first use.

Sequences batch automatically on 300M/600M (--batch_size, default 8): a padded, length-bucketed device forward per batch, masked so results are identical to running each sequence alone.

To embed a large batch faster, shard it across several cards with --devices 0,1,2,3: one worker per card, results reassembled in input order and identical to a single-card run:

tt-bio embed proteins.fasta --model esmc-600m --devices 0,1,2,3

Measured, not assumed: fanout only pays off when there's enough work per shard to amortize each worker's model-load and device-init cost. On small batches it can be flat or worse than a single card. esmc-6b scales to 4 cards on suitably large batches. See docs/esmc-multicard-scaling.md for the numbers before reaching for --devices on a small job.

For repeated/production embedding, submit to a persistent pool instead: a worker loads its model once and keeps it resident across every call, so the reload cost above is paid once per worker, not once per invocation:

tt-bio controller --listen 8765          # starts + keeps a worker per local card
tt-bio embed proteins.fasta --model esmc-6b --controller http://localhost:8765

The same capability is available from Python:

from tt_bio import esmc

emb = esmc.embed("MQIFVKTLTGKTITLEV...", model="esmc-600m")[0]
emb.per_residue   # [L, d_model] float32
emb.pooled        # [d_model] float32

# Shard a large set across cards (data-parallel, order preserved):
embs = esmc.embed(sequences, model="esmc-600m", devices=[0, 1, 2, 3])

Structure-Aware Protein Embeddings (SaProt)

SaProt is a structure-aware protein language model — an ESM-2 encoder over a fused amino-acid + Foldseek 3Di vocabulary (446 tokens). Where ESMC is sequence-only, SaProt also encodes local structure, so its embeddings and MLM logits reflect both sequence and shape. Use it for variant-effect / mutation-fitness scoring and function prediction when you have a structure (predicted or experimental — fold it with tt-bio predict first, then score it with SaProt).

tt-bio saprot proteins.fasta --model saprot-650m --structure structs/ --out_dir embeddings
tt-bio saprot proteins.fasta --model saprot-650m                # sequence-only (3Di = '#')

--structure is a PDB/cif file (single sequence) or a directory of <id>.pdb/<id>.cif files, one per FASTA id. The 3Di structural tokens are computed on host with Foldseek (conda install -c bioconda foldseek, or set FOLDSEEK_BIN); it runs off-device. Omit --structure for sequence-only mode (lower accuracy for 35M/650M; the 1.3B works sequence-only).

For each sequence you get per-residue structure-aware embeddings ([length, d_model] float32) and a pooled vector, plus per-residue MLM logits ([length, 446] with --logits) over the fused vocabulary — the log-likelihoods used for zero-shot mutation scoring. Output layout matches tt-bio embed (<id>.npz / embeddings.parquet / manifest.json).

--model selects the variant (saprot-35m, saprot-650m, saprot-1.3b). Parity vs the reference HuggingFace checkpoint and warm throughput are in docs/saprot-parity.md. The 35M port is deferred; use 650M.

Python:

from tt_bio import saprot

emb = saprot.embed(("MQIFVKTLTGKTITLEV...", "dweweaepvrdidi..."), model="saprot-650m")[0]
emb.per_residue   # [L, d_model] float32, structure-aware
emb.logits        # [L, 446] float32 (with return_logits=True)

Offline MSA (Optional)

Use this if you have enough disk and RAM and want local MSA. This avoids external MSA server calls and is faster for repeated runs.

tt-bio msa
tt-bio predict examples/prot.yaml --model boltz2 --override

tt-bio msa downloads UniRef30 to ~/.boltz/msa_db (~100GB download, ~500GB on disk after indexing). predict auto-detects this path.

To add EnvDB and use it in prediction: EnvDB can improve MSA coverage when UniRef30 hits are weak, at higher disk/RAM cost.

tt-bio msa --db all
tt-bio predict examples/prot.yaml --model boltz2 --use_envdb --override

Key Options:

• --override: Re-run from scratch, ignoring cached files
• --use_msa_server: Generate MSA via ColabFold API
• --msa_db_path: Use a local database at a custom path (e.g. --msa_db_path /data/colabfold_db)
• --use_envdb: Include EnvDB in offline MSA (tt-bio msa --db all)
• --accelerator=tenstorrent: Use Tenstorrent hardware (default, or use cpu/gpu)
• --fast: Makes some operations use a lower-precision numeric format that runs faster; accuracy is typically very close
• --debug: Show all raw output from the hardware and libraries instead of the progress display
• --debug --log: Same as --debug, but also print what each device is currently working on

Shared MSA Server (Optional)

Host the database on one machine and let others fetch MSAs from it over HTTP, so each prediction machine need not keep its own ~500GB copy.

# On the machine with the database:
tt-bio msa-server --listen 0.0.0.0:8765

# On any other machine (no local database needed):
tt-bio predict examples/prot.yaml --model protenix-v2 --msa_endpoint http://HOST:8765

The server runs the same offline colabfold_search and serves unpaired {hash}.a3m, with a shared cache and a search-concurrency cap (--max_concurrent). Add --token to require Authorization: Bearer <token>. --msa_endpoint applies to --model esmfold2/protenix-v2.

Binding Affinity Prediction (Boltz-2)

Predict binding affinity for protein-ligand complexes:

tt-bio predict examples/affinity.yaml --model boltz2 --use_msa_server --override --affinity_mw_correction

The --affinity_mw_correction flag applies molecular weight correction for more accurate predictions.

Input Format

ESMFold2 accepts protein inputs only. Protenix-v2 accepts proteins, DNA, RNA, ligands, and covalent bond constraints. OpenDDE currently accepts proteins only. Boltz-2 additionally supports affinity, pocket/contact constraints, potentials, and user-supplied templates.

Create a YAML file describing your complex:

version: 1
sequences:
  - protein:
      id: A
      sequence: MVTPEGNVSLVDESLLVGVTDEDRAVRSAHQFYERLIGLWAPAVMEAAHELGVFAALAEAPADSGELARRLDCDARAMRVLLDALYAYDVIDRIHDTNGFRYLLSAEARECLLPGTLFSLVGKFMHDINVAWPAWRNLAEVVRHGARDTSGAESPNGIAQEDYESLVGGINFWAPPIVTTLSRKLRASGRSGDATASVLDVGCGTGLYSQLLLREFPRWTATGLDVERIATLANAQALRLGVEERFATRAGDFWRGGWGTGYDLVLFANIFHLQTPASAVRLMRHAAACLAPDGLVAVVDQIVDADREPKTPQDRFALLFAASMTNTGGGDAYTFQEYEEWFTAAGLQRIETLDTPMHRILLARRATEPSAVPEGQASENLYFQ
  - ligand:
      id: B
      smiles: 'N[C@@H](Cc1ccc(O)cc1)C(=O)O'
properties:
  - affinity:
      binder: B

Entity Types:

• Polymers (protein, dna, rna): provide sequence
• Ligands (ligand): provide smiles or ccd code

Multiple Identical Chains:

- protein:
    id: [A, B]  # Two identical chains
    sequence: ...

Understanding Results

Output Structure

boltz_results_prot/
├── structures/
│   ├── prot.cif                      # Best-ranked predicted structure
│   └── prot_model_1.cif              # Additional samples (if diffusion_samples > 1)
├── results.json                      # One entry per target with confidence/affinity metrics
├── power_profile.csv                 # (optional, --report-energy)
├── power_profile.png                 # (optional, --report-energy)
├── prot_pae.npz                      # (optional, --write_pae)
├── prot_pde.npz                      # (optional, --write_pde)
└── prot_embeddings.npz               # (optional, --write_embeddings)

MSA results are cached in <out_dir>/msa/ (default ./msa/), keyed by sequence hash. The same protein sequence is never searched twice, even across different input files or runs. The MSA search uses all available CPU threads and keeps the database index memory-mapped for maximum speed.

Confidence Scores

Each target entry in results.json contains confidence metrics. The fields below are Boltz-2's; Protenix-v2 reports the same confidence_score / ptm / iptm / plddt (and all_runs when --diffusion_samples > 1, ranked best-first), while an ESMFold2 entry instead carries plddt (mean, 0-1), ptm when available, and n_residues / n_chains.

{
    "id": "prot",
    "status": "ok",
    "confidence_score": 0.84,
    "ptm": 0.84,
    "iptm": 0.82,
    "complex_plddt": 0.84,
    "chains_ptm": {
        "0": 0.85,
        "1": 0.83
    },
    "pair_chains_iptm": {
        "0": {"0": 0.85, "1": 0.72},
        "1": {"0": 0.82, "1": 0.83}
    }
}

• confidence_score: Overall confidence (0-1, higher is better), calculated as 0.8 × complex_plddt + 0.2 × iptm. Models are ranked by this score
• ptm: Predicted TM-score for complex (0-1)
• iptm: Interface TM-score (0-1)
• complex_plddt: Average per-residue confidence (0-1)
• chains_ptm: Per-chain TM-scores (0-1)
• pair_chains_iptm: Per-chain-pair interface TM-scores (0-1)

Affinity Predictions

For affinity targets, the same results.json entry also contains:

{
    "affinity_pred_value": 2.47,
    "affinity_probability_binary": 0.41,
    "affinity_pred_value1": 2.55,
    "affinity_pred_value2": 2.19,
    "affinity_probability_binary1": 0.50,
    "affinity_probability_binary2": 0.42
}

• affinity_probability_binary: Probability of binding (0-1). Use for hit discovery (higher = more likely to bind)
• affinity_pred_value: Predicted binding affinity as log10(IC50) in μM. Use for ligand optimization (lower = stronger binding). Only compare between known active molecules
• affinity_pred_value1, affinity_pred_value2: Individual model predictions for binding affinity
• affinity_probability_binary1, affinity_probability_binary2: Individual model predictions for binding probability

Advanced Usage

Input Format Details

Proteins with Custom MSA

- protein:
    id: A
    sequence: MVTPEGNVSLVDES...
    msa: ./path/to/msa.a3m

Proteins with Modifications

- protein:
    id: A
    sequence: MVTPEGNVSLVDES...
    modifications:
      - position: 5
        ccd: PTR  # Modified residue code

Ligands

- ligand:
    id: B
    smiles: 'CC1=CC=CC=C1'  # SMILES string
    # OR
    ccd: ATP                # CCD code

Constraints

Pocket and contact constraints are Boltz-2 only (they need a trained constraint embedder). Covalent bond constraints work with Boltz-2 and Protenix-v2.

Pocket Constraints (binding site):

constraints:
  - pocket:
      binder: B              # Ligand chain
      contacts: [[A, 10], [A, 11], [A, 12]]  # Binding site residues
      max_distance: 6.0      # Angstroms (4-20A, default 6A)
      force: false           # Use potential to enforce (default: false)

Contact Constraints:

constraints:
  - contact:
      token1: [A, 10]
      token2: [A, 50]
      max_distance: 8.0
      force: false

Bond Constraints (covalent link, e.g. a covalent inhibitor, glycosylation, or disulfide; works with Boltz-2 and Protenix-v2):

constraints:
  - bond:
      atom1: [A, 10, SG]     # [chain, residue, atom]
      atom2: [B, 1, C12]     # ligand atom by name; polymer atoms by residue

Templates

Use experimental structures as templates:

templates:
  - cif: ./template.cif
    chain_id: A
    template_id: A
    force: true              # Enforce template alignment
    threshold: 2.0           # Max deviation in Angstroms

Command-Line Options

Model-specific options are labelled below.

Common Options:

Affinity-Specific Options (Boltz-2):

MSA Options (Boltz-2, Protenix-v2, and OpenDDE use an MSA by default; ESMFold2 only when requested):

MSA Database Setup Options:

MSA Server Authentication

For --use_msa_server:

Basic Authentication:

export BOLTZ_MSA_USERNAME=myuser
export BOLTZ_MSA_PASSWORD=mypassword
tt-bio predict ... --model boltz2 --use_msa_server

API Key Authentication:

export MSA_API_KEY_VALUE=your-api-key
tt-bio predict ... --model boltz2 --use_msa_server

Optional: Multi-Machine Prediction

Combine the cards across any mix of Tenstorrent machines (a workstation, one or more QuietBoxes, one or more Galaxy servers) into a single run.

On the machine driving the run:

tt-bio predict ./proteins --model boltz2 --listen 8765 --use_msa_server --fast

On every additional machine, replace HOST with the driving machine's hostname or IP:

tt-bio worker --connect http://HOST:8765

Optional: Energy Measurement (Boltz-2)

Use --report-energy to profile energy during prediction:

tt-bio predict examples/686.yaml --model boltz2 --override --device_ids 0 --report-energy --energy-metric both --energy-sample-hz 5

Behavior:

• Select metric channel(s) with --energy-metric (tdp, input, both)
• Uses one sampling rate (--energy-sample-hz, default 20 Hz)
• Supports only Tenstorrent runs with one selected device
• Records two power channels when available:
  • power_w: tt-mgmt UMD telemetry power (TDP channel)
  • input_power_w: tt-mgmt UMD telemetry input power
• Requires optional tt-mgmt installation:
  • git clone --recursive https://github.com/aperezvicente-TT/tt-mgmt.git
  • pip install -e ./tt-mgmt
• Prints energy summary metrics for selected channels
• Always writes:
  • power_profile.csv
  • power_profile.png

BoltzGen

BoltzGen designs protein binders against a target. The pipeline runs design → inverse folding → folding → analysis → filtering and writes the top-ranked binders to <output>/final_ranked_designs/.

tt-bio gen run examples/binder.yaml --num_designs 10

This automatically uses every available card (splitting the designs across them and merging the results) and writes to ./binder/. Add --device_ids 0,2 to run on specific cards only.

Input Format

entities:
  - protein:
      id: B
      sequence: 80..120         # designed chain, sampled length per design
  - file:
      path: target.cif          # target structure (path relative to this yaml)
      include:
        - chain:
            id: A

80..120 randomises the binder length per design; a fixed integer pins it. Ligand, DNA, and RNA targets use the same YAML grammar as tt-bio predict. See the BoltzGen examples for binding sites, scaffolds, and residue constraints.

Protocols

--protocol sets defaults appropriate for the binder type.

Running a Subset

--steps restricts the pipeline.

tt-bio gen run examples/binder.yaml --steps design --num_designs 10
tt-bio gen run examples/binder.yaml --output existing/ --steps analysis filtering

Command-Line Options

Sequence Design (ProteinMPNN)

ProteinMPNN designs the amino-acid sequence most likely to fold into a fixed backbone (inverse folding) — the step run after a backbone generator like BoltzGen, or any bring-your-own-backbone PDB.

tt-bio design backbone.pdb --num-sequences 8 --checkpoint v_48_020.pt

Runs on CPU; a data-parallel fanout of processes scales throughput near-linearly (see docs/proteinmpnn-port.md). Writes <name>.fasta to --out-dir.

Command-Line Options

Cite

If you use this code or the models in your research, please cite the following papers:

@article{passaro2025boltz2,
  author = {Passaro, Saro and Corso, Gabriele and Wohlwend, Jeremy and Reveiz, Mateo and Thaler, Stephan and Somnath, Vignesh Ram and Getz, Noah and Portnoi, Tally and Roy, Julien and Stark, Hannes and Kwabi-Addo, David and Beaini, Dominique and Jaakkola, Tommi and Barzilay, Regina},
  title = {Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction},
  year = {2025},
  doi = {10.1101/2025.06.14.659707},
  journal = {bioRxiv}
}

@article{stark2025boltzgen,
  author = {Stark, Hannes and Faltings, Felix and Choi, MinGyu and Xie, Yuxin and Hur, Eunsu and O'Donnell, Timothy John and Bushuiev, Anton and U{\c c}ar, Talip and Passaro, Saro and Mao, Weian and Reveiz, Mateo and Bushuiev, Roman and Pluskal, Tom{\'a}{\v s} and Sivic, Josef and Kreis, Karsten and Vahdat, Arash and Ray, Shamayeeta and Goldstein, Jonathan T. and Savinov, Andrew and Hambalek, Jacob A. and Gupta, Anshika and Taquiri-Diaz, Diego A. and Zhang, Yaotian and Hatstat, A. Katherine and Arada, Angelika and Kim, Nam Hyeong and Tackie-Yarboi, Ethel and Boselli, Dylan and Schnaider, Lee and Liu, Chang C. and Li, Gene-Wei and Hnisz, Denes and Sabatini, David M. and DeGrado, William F. and Wohlwend, Jeremy and Corso, Gabriele and Barzilay, Regina and Jaakkola, Tommi},
  title = {BoltzGen: Toward Universal Binder Design},
  year = {2025},
  doi = {10.1101/2025.11.20.689494},
  journal = {bioRxiv}
}

@article{wohlwend2024boltz1,
  author = {Wohlwend, Jeremy and Corso, Gabriele and Passaro, Saro and Getz, Noah and Reveiz, Mateo and Leidal, Ken and Swiderski, Wojtek and Atkinson, Liam and Portnoi, Tally and Chinn, Itamar and Silterra, Jacob and Jaakkola, Tommi and Barzilay, Regina},
  title = {Boltz-1: Democratizing Biomolecular Interaction Modeling},
  year = {2024},
  doi = {10.1101/2024.11.19.624167},
  journal = {bioRxiv}
}

@misc{candido2026language,
  author = {Candido, Salvatore and Hayes, Thomas and Derry, Alexander and Rao, Roshan and Lin, Zeming and Verkuil, Robert and others},
  title = {Language Modeling Materializes a World Model of Protein Biology},
  year = {2026},
  url = {https://biohub.ai/papers/esm_protein.pdf},
  note = {Preprint; ESMC / ESMFold2}
}

@misc{protenix2025,
  author = {{ByteDance AML AI4Science Team}},
  title = {Protenix: An AlphaFold3 Reproduction for Biomolecular Structure Prediction},
  year = {2025},
  url = {https://github.com/bytedance/Protenix}
}

In addition if you use the automatic MSA generation, please cite:

@article{mirdita2022colabfold,
  title={ColabFold: making protein folding accessible to all},
  author={Mirdita, Milot and Sch{\"u}tze, Konstantin and Moriwaki, Yoshitaka and Heo, Lim and Ovchinnikov, Sergey and Steinegger, Martin},
  journal={Nature methods},
  year={2022}
}

License

tt-bio is released under the MIT License (see LICENSE) and is built on the MIT-licensed Boltz-2 / Boltz-1 code. It bundles third-party code, each under its upstream license: the ESMFold2 host-side reference under tt_bio/_vendor/ (the esm pipeline, MIT, © Chan Zuckerberg Biohub; and the HuggingFace ESMFold2 model definition, Apache-2.0) and the BoltzGen binder-design source under tt_bio/boltzgen/ (MIT, © Hannes Stärk). Protenix-v2 is an independent ttnn reimplementation (no upstream code is vendored) and its weights download from ByteDance's Hugging Face mirror under Apache-2.0. See NOTICE for sources, versions, and modifications.