File size: 8,311 Bytes
e2bfccc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 | # Directory Navigation Guide
This guide maps the project by responsibility. Use it when a new thread needs to find the SSM code, LLM wrapper, data pipeline, train/test scripts, or remote-run artifacts quickly.
## Local Workspace
Current local workspace root:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern
```
Main local repos:
| Purpose | Local path | GitHub |
|---|---|---|
| Experiment ledger | `C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\Taotern_LLM_Experiments` | `https://github.com/StarMists/Taotern_LLM_Experiments` |
| SSM model | `C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\Taotern_SSM` | `https://github.com/StarMists/gamma_SSM_S4_enhanced` |
| TaoTrain LLM code | `C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\TaoTrain` | `https://github.com/lobakkang/TaoTrain` |
| Remote run tool | `C:\Users\YouZheng\Documents\LYZ\MyContent\MyComp\RepoBridge` | local tool repo |
| TaoData scripts | not currently cloned under this workspace | `https://github.com/lobakkang/TaoData` |
## Experiment Ledger
Path:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\Taotern_LLM_Experiments
```
Important files:
| File | Purpose |
|---|---|
| `README.md` | Current SSM LLM status and attention TaoNet comparison |
| `experiments/index.csv` | Searchable run ledger |
| `experiments/runs/<run_id>/manifest.yaml` | Run purpose, commits, data, status |
| `experiments/runs/<run_id>/summary.md` | Human-readable result |
| `experiments/runs/<run_id>/metrics.csv` | Compact metric snapshot |
| `experiments/runs/<run_id>/repobridge.config.json` | Exact remote-run config |
| `experiments/resources/tokenizers/` | Small tokenizer configs only |
| `docs/WORKFLOW.md` | How future runs should be recorded |
| `docs/CURRENT_SSM_LLM_ARCHITECTURE.md` | Current TaoNet-SSM layers, equations, matrices, parameters |
Rule: keep this repo compact. Commit summaries/configs/CSV metrics, not raw output trees or checkpoints.
## SSM Model Repo
Path:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\Taotern_SSM
```
Main code locations:
| Area | File or directory | Notes |
|---|---|---|
| DPLR SSM core | `gamma_space_model/modules/s4_ternary_dplr_ssm.py` | Current main SSM core for TaoNet-SSM |
| Gamma S4 core | `gamma_space_model/modules/ssm_gamma_s4.py` | Older Gamma/S4-style core |
| Baseline Gamma core | `gamma_space_model/modules/ssm_gamma.py` | Baseline/reference SSM |
| SSM blocks | `gamma_space_model/modules/block*.py` | Standalone SSM block wrappers |
| TileLang/Triton fallback area | `csrc/tilelang/` | Capability detection and fallback code |
| Selective scan op wrapper | `gamma_space_model/ops/selective_scan_interface.py` | SSM op interface |
| DPLR profiler | `scripts/profile_dplr_frequency_path.py` | Profiles DPLR frequency path |
| TileLang diagnosis | `scripts/diagnose_tilelang_acceleration.py` | Reports real vs fallback acceleration |
| SSM variant benchmark | `scripts/benchmark_ssm_variants.py` | Standalone SSM benchmarks |
| SSM tests | `tests/test_s4_ternary_dplr_ssm.py`, `tests/test_ssm_gamma*.py` | Core correctness tests |
| Historical record | `EXPERIMENT_RECORD.md` | Older narrative record; new LLM records should be mirrored into this experiment ledger |
When improving the SSM model itself, start from:
```text
gamma_space_model/modules/s4_ternary_dplr_ssm.py
```
When working on hardware acceleration, start from:
```text
csrc/tilelang/
scripts/profile_dplr_frequency_path.py
scripts/diagnose_tilelang_acceleration.py
```
Remote SSM path used by RepoBridge runs:
```text
/home/student/YouZheng/gamma_ssm_repo
```
## TaoTrain LLM Repo
Path:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\TaoTrain
```
Main code locations:
| Area | File or directory | Notes |
|---|---|---|
| Attention TaoNet baseline | `src/taoTrain/models/taonet.py` | Reference model for comparisons |
| SSM TaoNet wrapper | `src/taoTrain/models/taonet_ssm.py` | Replaces attention core with SSM mixer |
| Model config schema | `src/taoTrain/config.py` | SSM flags live here: hidden dim, mixer dim, shift, kernel mode |
| Model registry | `src/taoTrain/models/registry.py` | Architecture registration |
| Token/data utilities | `src/taoTrain/data/` | JSONL and tokenization data paths |
| Tokenizer trainer | `src/taoTrain/tokenizers/trainer.py` | SentencePiece training path |
| Training loop | `src/taoTrain/training/trainer.py` | Full trainer implementation |
| CLI | `src/taoTrain/cli.py` | TaoTrain command entry |
| Real-token benchmark | `scripts/benchmark_taonet_real_tokens.py` | Main attention vs SSM benchmark for TaoData token tasks |
| Synthetic token benchmark | `scripts/benchmark_taonet_token_variants.py` | Previous/increment/random token probes |
| TaoData pilot tokenizer config | `configs/tokenizer_taodata_pilot.yaml` | Generated pilot 8k SentencePiece tokenizer |
| SSM pretrain config | `configs/ssm_pretrain.yaml` | Config path for SSM pretraining experiments |
| SSM wrapper tests | `tests/test_taonet_ssm.py` | Shape and config behavior tests |
Current real-token benchmark entry point:
```text
scripts/benchmark_taonet_real_tokens.py
```
Current SSM wrapper entry point:
```text
src/taoTrain/models/taonet_ssm.py
```
Current attention baseline:
```text
src/taoTrain/models/taonet.py
```
Remote TaoTrain path used by RepoBridge:
```text
/home/student/YouZheng/repo
```
## TaoData
GitHub:
```text
https://github.com/lobakkang/TaoData
```
Current local status:
```text
No local TaoData checkout was found under C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase as of 2026-04-30.
```
Remote data path used in current benchmarks:
```text
/home/student/Data/TaoData/pretrain.jsonl.fineweb.jsonl
```
Current pilot tokenizer path on remote:
```text
/home/student/YouZheng/tokenizers/taodata_pilot_8k/tokenizer.model
/home/student/YouZheng/tokenizers/taodata_pilot_8k/tokenizer.vocab
```
Tokenizer config snapshot in this ledger:
```text
experiments/resources/tokenizers/taodata_pilot_8k.yaml
```
When TaoData is cloned locally, update this guide with the exact data download/generation scripts and any preprocessing entry points.
## RepoBridge
Path:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyComp\RepoBridge
```
Important files:
| File or directory | Purpose |
|---|---|
| `repobridge/core.py` | Sync, SSH, SFTP, run, download implementation |
| `repobridge/cli.py` | CLI entry point |
| `repobridge/app.py` | GUI |
| `CODEX_OPERATOR_GUIDE.md` | Codex remote-run guide |
| `PRODUCTION_RUNBOOK.md` | Production checklist |
| old `repobridge.*.config.json` files | Historical configs; new experiment configs should live in this ledger |
Preferred future location for experiment configs:
```text
Taotern_LLM_Experiments\experiments\runs\<run_id>\repobridge.config.json
```
Remote write root:
```text
/home/student/YouZheng
```
Remote output base:
```text
/home/student/YouZheng/outputs-taotrain
```
Important operational note:
Avoid downloading the whole remote output base if it contains many historical runs. Prefer downloading or copying only the specific run folder.
## Current Best SSM LLM Path
To inspect the current best SSM LLM implementation:
1. Open TaoTrain wrapper:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\TaoTrain\src\taoTrain\models\taonet_ssm.py
```
2. Follow the DPLR core import into:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\Taotern_SSM\gamma_space_model\modules\s4_ternary_dplr_ssm.py
```
3. Compare against attention TaoNet:
```text
C:\Users\YouZheng\Documents\LYZ\MyContent\MyLLM\Codebase\Taotern\TaoTrain\src\taoTrain\models\taonet.py
```
4. Reproduce current best benchmark with:
```text
Taotern_LLM_Experiments\experiments\runs\2026-04-29_spm_b32_500step_mixer_sweep\repobridge.config.json
```
5. Read current conclusion in:
```text
Taotern_LLM_Experiments\README.md
```
|