main <- step 6000

README.md

@@ -0,0 +1,19 @@
+---
+license: apache-2.0
+library_name: pytorch
+tags: [dynamic-chunking, tokenization, hnet, byte-level, beetle-hnet, legacy-trainer]
+---
+
+# pilot_D_s0
+
+Byte-level H-Net run from [beetle-hnet](https://github.com/suchirsalhan/beetle-hnet), trained
+with the legacy trainer (`legacy/tier1/train.py`). Split out of the bundled repo
+[`AdaptiveChunking/hnet-chunking-pilots`](https://huggingface.co/AdaptiveChunking/hnet-chunking-pilots) into the Beetle per-run layout.
+
+- **`step-N` branches**: 1 retained checkpoints (6000),
+  each with `model.safetensors` and `config.json` (`model_config`, training config and the
+  `tied_weights` map: restore with `sd[dst] = sd[src]` for every entry).
+- **`main`**: the latest weights plus `final.json` and `log.jsonl`.
+- The legacy trainer kept no optimiser state, so there is no `training_state/` here. FLORES+
+  boundary probes for this run are in the dataset
+  [`AdaptiveChunking/hnet-chunking-probes`](https://huggingface.co/datasets/AdaptiveChunking/hnet-chunking-probes) under `pilot_D_s0/`.

config.json

@@ -0,0 +1,109 @@
+{
+  "run": "pilot_D_s0",
+  "step": 6000,
+  "model_config": {
+    "vocab_size": 256,
+    "d_enc": 256,
+    "d_main": 512,
+    "n_enc": 2,
+    "n_dec": 2,
+    "n_main": 6,
+    "n_heads_enc": 4,
+    "n_heads_main": 8,
+    "seq_len": 1024,
+    "dropout": 0.0,
+    "target_ratio": 6.0,
+    "max_chunk_mult": 3.0,
+    "dechunk": "ema",
+    "router_noise": 0.0,
+    "rope_base": 10000.0
+  },
+  "tied_weights": {
+    "head.weight": "emb.weight"
+  },
+  "train_config": {
+    "condition": "D",
+    "seed": 0,
+    "out_dir": null,
+    "run_name": "pilot_D_s0",
+    "corpus": "corpus",
+    "langs": null,
+    "seq_len": 1024,
+    "batch_size": 32,
+    "grad_accum": 1,
+    "steps": 6000,
+    "lr": 0.0003,
+    "min_lr_frac": 0.1,
+    "warmup": 300,
+    "weight_decay": 0.1,
+    "grad_clip": 1.0,
+    "betas": [
+      0.9,
+      0.95
+    ],
+    "router_lr_mult": 1.0,
+    "d_enc": 256,
+    "d_main": 512,
+    "n_enc": 2,
+    "n_dec": 2,
+    "n_main": 6,
+    "n_heads_enc": 4,
+    "n_heads_main": 8,
+    "dechunk": "ema",
+    "target_ratio": 6.0,
+    "alpha": 0.03,
+    "alpha_parity_cap": 0.06,
+    "global_anchor": 0.0,
+    "premium_clip": 4.0,
+    "premium_kind": "flores_per_sent",
+    "mixture": null,
+    "doc_filters": null,
+    "tau1": 0.4,
+    "tau2": 0.7,
+    "freeze_scope": "router",
+    "ckpt_every": 200,
+    "eval_every": 200,
+    "log_every": 50,
+    "probe_sents": 128,
+    "eval_batches": 8,
+    "keep_ckpt": false,
+    "amp": "bf16",
+    "compile": false,
+    "cpc_decay": 0.9,
+    "device": "cuda",
+    "log_spaced_probe": true,
+    "probe_bpb_batches": 4,
+    "dump_reps": true,
+    "dump_reps_thresholds": [
+      0.3,
+      0.4,
+      0.5,
+      0.6,
+      0.7
+    ],
+    "router_noise": 0.0,
+    "genreg_target_entropy": null,
+    "genreg_eta": 0.05
+  },
+  "probe_agg": {
+    "gini_cpc": 0.2082899216246239,
+    "gini_cps": 0.07551111358698681,
+    "bpic": 10.689683026015263,
+    "worst_lang": "zh",
+    "cpc_parity_vs_en": {
+      "en": 1.0,
+      "de": 0.8811880422902347,
+      "vi": 0.8550701968093136,
+      "tr": 0.9052453060798022,
+      "fi": 0.8773813386702968,
+      "ru": 0.8396012659981589,
+      "el": 0.7153865165706734,
+      "hi": 0.8831835196901316,
+      "ta": 0.6167014530058299,
+      "th": 0.8654255089556684,
+      "zh": 2.785062043583627,
+      "my": 0.5680097516780557
+    }
+  },
+  "source": "AdaptiveChunking/hnet-chunking-pilots/pilot_D_s0/model.safetensors"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7a846a04ad851b19efc1eb026cbb344e6b3419049d0d8cb0364b613c5dbe9f6
+size 90078296