Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75899.45532405979,
-      "required_pages_per_second_upper_estimate": 0.006578772422078673
+      "required_token_positions_per_second": 75632.4423830858,
+      "required_pages_per_second_upper_estimate": 0.00655562841709316
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9487.431915507474,
-      "required_pages_per_second_upper_estimate": 0.003784678440843894
+      "required_token_positions_per_second": 9454.055297885725,
+      "required_pages_per_second_upper_estimate": 0.0037713640090496748
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18974.863831014947,
-      "required_pages_per_second_upper_estimate": 0.004389500801107561
+      "required_token_positions_per_second": 18908.11059577145,
+      "required_pages_per_second_upper_estimate": 0.0043740586149507935
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28462.295746522417,
-      "required_pages_per_second_upper_estimate": 0.0017332044808382734
+      "required_token_positions_per_second": 28362.165893657173,
+      "required_pages_per_second_upper_estimate": 0.0017271070981395174
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28462.295746522417,
-      "required_pages_per_second_upper_estimate": 0.0021911063152446216
+      "required_token_positions_per_second": 28362.165893657173,
+      "required_pages_per_second_upper_estimate": 0.0021833980419938824
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28462.295746522417,
-      "required_pages_per_second_upper_estimate": 0.35340645599565934
+      "required_token_positions_per_second": 28362.165893657173,
+      "required_pages_per_second_upper_estimate": 0.35216317833613336
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08713245738275077,
-    "producer_p95_seconds": 0.28750059899903135,
-    "gpu_update_seconds": 0.9713903701312671,
-    "producer_batches_per_second": 11.47677949225343,
-    "gpu_batches_per_second": 1.0294522477764185,
-    "producer_headroom": 11.148433078894994,
+    "producer_batch_seconds": 0.15434361933614582,
+    "producer_p95_seconds": 0.30774384700271185,
+    "gpu_update_seconds": 0.9748197688309521,
+    "producer_batches_per_second": 6.479049825973657,
+    "gpu_batches_per_second": 1.025830652982392,
+    "producer_headroom": 6.315905853599861,
     "sustainable": true,
-    "buffer_seconds": 60.22620294813856,
+    "buffer_seconds": 60.43882566751903,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 248590.9986309057,
-    "upload_bytes_per_second": 5086383.092773639,
+    "generation_bytes_per_second": 256317.55874757678,
+    "upload_bytes_per_second": 4995265.12301872,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.34184140391711,
+    "estimated_full_shards_per_hour": 54.99978133984067,
     "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": 1.0379702050013293,
-      "bytes": 6386142,
-      "pages": 34,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1697385600018606,
-      "bytes": 11844982,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.079595678998885,
       "bytes": 1118845,
@@ -562,6 +550,18 @@
       "bytes": 11727417,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.4179154230005224,
+      "bytes": 11130810,
+      "pages": 34,
+      "shards": 4
+    },
+    {
+      "seconds": 0.9252583010020317,
+      "bytes": 184838,
+      "pages": 5,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 379800, "loss": 0.5833548456430435, "lr": 0.0003, "grad_norm": 0.1605502428370705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55591490906, "tokens_per_second": 150128.03480669996, "data_wait_s": 8.499999239575118e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 379900, "loss": 0.595781497657299, "lr": 0.0003, "grad_norm": 0.1792857469272972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55606176051, "tokens_per_second": 149174.9570351013, "data_wait_s": 1.2386997696012259e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380000, "loss": 0.5973047744482756, "lr": 0.0003, "grad_norm": 0.14794959448736272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55620842514, "tokens_per_second": 149050.32637906037, "data_wait_s": 1.7037000361597165e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 380100, "loss": 0.6261854413896799, "lr": 0.0003, "grad_norm": 0.18086174655739745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55635464482, "tokens_per_second": 134259.735991459, "data_wait_s": 1.5268000424839556e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1574e29d847fd36517b981709061584b340ca148d65d46b98c49745ef977d7d5
-size 533580
+oid sha256:f44db8a92ff598032b2200a0873d467bbf1c088c77e3cd2ce2a940e683638568
+size 535698