---
license: other
license_name: mixed
pretty_name: FlySLM run-time data
tags:
- connectome
- drosophila
- spiking-neural-network
---

# FlySLM run-time data

Everything the FlySLM app downloads after installation, mirrored in one place so that a moved or changed upstream
link can never break an installation: the fly-brain wiring data (connectomes), the conversation-data sources, the
local helper and translator models, the llama.cpp runtime for every platform and the pre-trained Fly AI brains.

FlySLM pins a commit of this dataset and verifies every file by size and SHA-256 before using it
(`manifest.json` lists them). The original source of every file is recorded below; FlySLM never downloads from there.

**Licences differ per file and are the original ones** (this mirror changes nothing). Some conversation sources are
**non-commercial** (DailyDialog CC-BY-NC-SA-4.0, EmpatheticDialogues CC-BY-NC-4.0); databricks-dolly-15k is
CC-BY-SA-3.0 and TinyStories CDLA-Sharing-1.0 (share-alike). The connectomes are CC-BY 4.0: cite the papers below.

## Layout

| Folder | Contents |
|---|---|
| `helper/` | Gemma 4 12B (QAT, 4-bit GGUF) and its vision projector: the local AI Helper |
| `translator/` | Hy-MT2 1.8B (6-bit GGUF): the local translator |
| `runtime/llama.cpp-b11146/` | unmodified official llama.cpp release archives (Windows, Linux, macOS) |
| `connectomes/` | raw FlyWire FAFB v783 and MaleCNS v1.0 files, as published |
| `corpus/` | the conversation and text sources FlySLM builds its training dataset from, as published |
| `brains/` | pre-trained Fly AI brains (FlySLM/m and FlySLM/f) with their training records |

## Files

| Path | Size | Licence | Original source |
|---|---|---|---|
| `helper/gemma-4-12b-it-qat-q4_0/gemma-4-12b-it-qat-q4_0.gguf` | 6,652.7 MB | Apache-2.0 | https://huggingface.co/google/gemma-4-12b-it-qat-q4_0-gguf @ 29d097773436b69ff9feafd636ab4cf873786537 |
| `helper/gemma-4-12b-it-qat-q4_0/mmproj-gemma-4-12b-it-qat-q4_0.gguf` | 167.0 MB | Apache-2.0 | https://huggingface.co/google/gemma-4-12b-it-qat-q4_0-gguf @ 29d097773436b69ff9feafd636ab4cf873786537 |
| `translator/hy-mt2-1.8b/Hy-MT2-1.8B-Q6_K.gguf` | 1,406.5 MB | Apache-2.0 | https://huggingface.co/tencent/Hy-MT2-1.8B-GGUF @ a0c709d9fac510f2c807aa3af52872340dc37a4a |
| `runtime/llama.cpp-b11146/llama-b11146-bin-win-vulkan-x64.zip` | 30.6 MB | MIT | https://github.com/ggml-org/llama.cpp/releases/download/b11146/llama-b11146-bin-win-vulkan-x64.zip |
| `runtime/llama.cpp-b11146/llama-b11146-bin-win-cpu-x64.zip` | 17.7 MB | MIT | https://github.com/ggml-org/llama.cpp/releases/download/b11146/llama-b11146-bin-win-cpu-x64.zip |
| `runtime/llama.cpp-b11146/llama-b11146-bin-ubuntu-vulkan-x64.tar.gz` | 29.2 MB | MIT | https://github.com/ggml-org/llama.cpp/releases/download/b11146/llama-b11146-bin-ubuntu-vulkan-x64.tar.gz |
| `runtime/llama.cpp-b11146/llama-b11146-bin-ubuntu-x64.tar.gz` | 16.2 MB | MIT | https://github.com/ggml-org/llama.cpp/releases/download/b11146/llama-b11146-bin-ubuntu-x64.tar.gz |
| `runtime/llama.cpp-b11146/llama-b11146-bin-macos-arm64.tar.gz` | 10.7 MB | MIT | https://github.com/ggml-org/llama.cpp/releases/download/b11146/llama-b11146-bin-macos-arm64.tar.gz |
| `runtime/llama.cpp-b11146/llama-b11146-bin-macos-x64.tar.gz` | 10.7 MB | MIT | https://github.com/ggml-org/llama.cpp/releases/download/b11146/llama-b11146-bin-macos-x64.tar.gz |
| `connectomes/fafb_v783/Supplemental_file1_neuron_annotations.tsv` | 30.2 MB | CC-BY 4.0 | https://raw.githubusercontent.com/flyconnectome/flywire_annotations/17fc57722002e1a7d38cdd0c89ac382bf92718da/supplemental_files/Supplemental_file1_neuron_annotations.tsv |
| `connectomes/fafb_v783/proofread_connections_783.feather` | 812.6 MB | CC-BY 4.0 | https://zenodo.org/api/records/10676866/files/proofread_connections_783.feather/content |
| `connectomes/malecns_v1.0/body-annotations-male-cns-v1.0-minconf-0.5.feather` | 13.8 MB | CC-BY 4.0 | https://storage.googleapis.com/flyem-male-cns/v1.0/connectome-data/flat-connectome/body-annotations-male-cns-v1.0-minconf-0.5.feather |
| `connectomes/malecns_v1.0/body-neurotransmitters-male-cns-v1.0.feather` | 41.3 MB | CC-BY 4.0 | https://storage.googleapis.com/flyem-male-cns/v1.0/connectome-data/flat-connectome/body-neurotransmitters-male-cns-v1.0.feather |
| `connectomes/malecns_v1.0/connectome-weights-male-cns-v1.0-minconf-0.5-traced-only.feather` | 484.5 MB | CC-BY 4.0 | https://storage.googleapis.com/flyem-male-cns/v1.0/connectome-data/flat-connectome/connectome-weights-male-cns-v1.0-minconf-0.5-traced-only.feather |
| `corpus/everyday_conversations/train_sft.parquet` | 2.0 MB | Apache-2.0 | https://huggingface.co/datasets/HuggingFaceTB/everyday-conversations-llama3.1-2k/resolve/14f543216b9ba42b6b951dc5bd199460d193b162/data/train_sft-00000-of-00001.parquet |
| `corpus/everyday_conversations/test_sft.parquet` | 0.1 MB | Apache-2.0 | https://huggingface.co/datasets/HuggingFaceTB/everyday-conversations-llama3.1-2k/resolve/14f543216b9ba42b6b951dc5bd199460d193b162/data/test_sft-00000-of-00001.parquet |
| `corpus/dailydialog/train.zip` | 1.8 MB | CC-BY-NC-SA-4.0 | https://huggingface.co/datasets/roskoN/dailydialog/resolve/5214b2a66405abf87fd229e5c1007985501ffe3e/train.zip |
| `corpus/dailydialog/validation.zip` | 0.2 MB | CC-BY-NC-SA-4.0 | https://huggingface.co/datasets/roskoN/dailydialog/resolve/5214b2a66405abf87fd229e5c1007985501ffe3e/validation.zip |
| `corpus/dailydialog/test.zip` | 0.2 MB | CC-BY-NC-SA-4.0 | https://huggingface.co/datasets/roskoN/dailydialog/resolve/5214b2a66405abf87fd229e5c1007985501ffe3e/test.zip |
| `corpus/empathetic_dialogues/empatheticdialogues.tar.gz` | 26.7 MB | CC-BY-NC-4.0 | https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/empatheticdialogues.tar.gz |
| `corpus/soda/valid.parquet` | 79.0 MB | CC-BY-4.0 | https://huggingface.co/datasets/allenai/soda/resolve/fdc848ab0183208ea7808206c91c724414d0a071/valid.parquet |
| `corpus/oasst1/train.parquet` | 37.7 MB | Apache-2.0 | https://huggingface.co/datasets/OpenAssistant/oasst1/resolve/fdf72ae0827c1cda404aff25b6603abec9e3399b/data/train-00000-of-00001-b42a775f407cee45.parquet |
| `corpus/dolly/databricks-dolly-15k.jsonl` | 12.5 MB | CC-BY-SA-3.0 | https://huggingface.co/datasets/databricks/databricks-dolly-15k/resolve/bdd27f4d94b9c1f951818a7da7fd7aeea5dbff1a/databricks-dolly-15k.jsonl |
| `corpus/tinystories/validation.parquet` | 9.5 MB | CDLA-Sharing-1.0 | https://huggingface.co/datasets/roneneldan/TinyStories/resolve/f54c09fd23315a6f9c86f9dc80f725de7d8f9c64/data/validation-00000-of-00001-869c898b519ad725.parquet |
| `corpus/tinystories/train-00000-of-00004.parquet` | 237.2 MB | cdla-sharing-1.0 | https://huggingface.co/datasets/roneneldan/TinyStories/resolve/f54c09fd23315a6f9c86f9dc80f725de7d8f9c64/data/train-00000-of-00004-2d5a1467fff1081b.parquet |
| `brains/flyslm-m-full-thorough-2026-10-04/checkpoint/checkpoint.pt` | 134.1 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/checkpoint/metadata.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/tokenizer/provenance.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/tokenizer/tokenizer.json` | 0.1 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/package/edges.npz` | 20.6 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/package/manifest.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/package/neurons.parquet` | 4.8 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/package/regions.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/card.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-m-full-thorough-2026-10-04/run.json` | 0.1 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_m_full_20261004-154407) |
| `brains/flyslm-f-brain-thorough-2026-10-04/checkpoint/checkpoint.pt` | 117.2 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/checkpoint/metadata.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/tokenizer/provenance.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/tokenizer/tokenizer.json` | 0.1 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/package/edges.npz` | 9.2 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/package/manifest.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/package/neurons.parquet` | 4.1 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/package/regions.json` | 0.1 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/card.json` | 0.0 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |
| `brains/flyslm-f-brain-thorough-2026-10-04/run.json` | 0.1 MB | see card.json (connectome CC-BY 4.0; trained on data including non-commercial sources) | trained with FlySLM (run run_flyai_f_brain_20261004-154452) |

## Attribution

- Berg S. et al. Sexual dimorphism in the complete connectome of the Drosophila male central nervous system (MaleCNS v1.0, released 2026-06-08; Cell 2026). FlyEM (HHMI Janelia), University of Cambridge, MRC LMB and Google Research. Data: gs://flyem-male-cns/v1.0/.
- Conover M. et al. (2023) Free Dolly: Introducing the World's First Truly Open Instruction-Tuned LLM. Databricks.
- Dorkenwald S. et al. (2024) Neuronal wiring diagram of an adult brain. Nature 634:124-138. Schlegel P. et al. (2024) Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature 634:139-152. FlyWire Consortium, Zenodo doi:10.5281/zenodo.10676866. Neurotransmitter predictions: Eckstein N., Bates A.S. et al. (2024) Cell 187:2574.
- Eldan R., Li Y. (2023) TinyStories. arXiv:2305.07759.
- Eldan R., Li Y. (2023) TinyStories. arXiv:2305.07759. Licence CDLA-Sharing-1.0.
- HuggingFaceTB/everyday-conversations-llama3.1-2k (synthetic everyday chats generated with Llama 3.1).
- Kim H. et al. (2023) SODA: Million-scale Dialogue Distillation with Social Commonsense Contextualization. EMNLP. (allenai/soda)
- Köpf A. et al. (2023) OpenAssistant Conversations. NeurIPS Datasets and Benchmarks.
- Li Y., Su H., Shen X., Li W., Cao Z., Niu S. (2017) DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset. IJCNLP.
- Rashkin H., Smith E. M., Li M., Boureau Y-L. (2019) Towards Empathetic Open-domain Conversation Models. ACL.
- Gemma 4 (Google), Apache-2.0. Hy-MT2 (Tencent Hunyuan), Apache-2.0. llama.cpp (ggml-org), MIT.