Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76635.46852404454,
-      "required_pages_per_second_upper_estimate": 0.005301819531279024
+      "required_token_positions_per_second": 77349.37655460571,
+      "required_pages_per_second_upper_estimate": 0.005351209345328098
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9579.433565505567,
-      "required_pages_per_second_upper_estimate": 0.004010708792556877
+      "required_token_positions_per_second": 9668.672069325714,
+      "required_pages_per_second_upper_estimate": 0.004048071090594505
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19158.867131011135,
-      "required_pages_per_second_upper_estimate": 0.003990286770086561
+      "required_token_positions_per_second": 19337.344138651428,
+      "required_pages_per_second_upper_estimate": 0.004027458824022828
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28738.300696516704,
-      "required_pages_per_second_upper_estimate": 0.001722817245911796
+      "required_token_positions_per_second": 29006.01620797714,
+      "required_pages_per_second_upper_estimate": 0.0017388663820459324
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28738.300696516704,
-      "required_pages_per_second_upper_estimate": 0.0022342764727284526
+      "required_token_positions_per_second": 29006.01620797714,
+      "required_pages_per_second_upper_estimate": 0.0022550901761884145
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28738.300696516704,
-      "required_pages_per_second_upper_estimate": 0.3686240645517208
+      "required_token_positions_per_second": 29006.01620797714,
+      "required_pages_per_second_upper_estimate": 0.3720580316822147
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.19743127892186862,
-    "producer_p95_seconds": 0.3337966310027696,
-    "gpu_update_seconds": 0.9620610589321011,
-    "producer_batches_per_second": 5.06505354906676,
-    "gpu_batches_per_second": 1.039435065701559,
-    "producer_headroom": 4.872890780962964,
+    "producer_batch_seconds": 0.11064850195316467,
+    "producer_p95_seconds": 0.3272133220016258,
+    "gpu_update_seconds": 0.9531815676361767,
+    "producer_batches_per_second": 9.03762800533242,
+    "gpu_batches_per_second": 1.049118063077877,
+    "producer_headroom": 8.614500429835369,
     "sustainable": true,
-    "buffer_seconds": 60.60984671272237,
+    "buffer_seconds": 60.05043876107913,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 647942.418780718,
-    "upload_bytes_per_second": 6061072.4821805665,
+    "generation_bytes_per_second": 645014.9702228374,
+    "upload_bytes_per_second": 6030576.222049866,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.51667987139776,
+    "estimated_full_shards_per_hour": 69.20259871489449,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.186567714001285,
-      "bytes": 11920154,
-      "pages": 44,
-      "shards": 3
-    },
     {
       "seconds": 0.9970581140005379,
       "bytes": 6389566,
@@ -562,6 +556,12 @@
       "bytes": 7362354,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.2885511929998756,
+      "bytes": 10359776,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 515300, "loss": 0.6192801594734192, "lr": 0.0003, "grad_norm": 0.14641507097440298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75422702197, "tokens_per_second": 151600.634916967, "data_wait_s": 1.7659000150160864e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 515400, "loss": 0.6087738871574402, "lr": 0.0003, "grad_norm": 0.15191230004144576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75437139886, "tokens_per_second": 150641.78121673112, "data_wait_s": 2.0651001250371337e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 515500, "loss": 0.5553947687149048, "lr": 0.0003, "grad_norm": 0.14193700794817124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75451684080, "tokens_per_second": 151055.89318762525, "data_wait_s": 1.1118998372694477e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 515600, "loss": 0.5998923778533936, "lr": 0.0003, "grad_norm": 0.16084814012242907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75466232709, "tokens_per_second": 151145.74336640813, "data_wait_s": 1.04649989225436e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4285646f9ff7e73186c3de78d69cd610dfaacedd139109a23e838800349ad268
-size 650853
+oid sha256:d88bc6a28eec9b132bc3efdd6d4468e53e9113754991768359b147e53ed0a028
+size 652971