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": 77156.25604534235,
-      "required_pages_per_second_upper_estimate": 0.0053378488203959405
+      "required_token_positions_per_second": 76250.24128246879,
+      "required_pages_per_second_upper_estimate": 0.005275168616856465
     },
     {
       "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": 9644.532005667794,
-      "required_pages_per_second_upper_estimate": 0.004037964150042786
+      "required_token_positions_per_second": 9531.280160308599,
+      "required_pages_per_second_upper_estimate": 0.003990547967358352
     },
     {
       "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": 19289.064011335588,
-      "required_pages_per_second_upper_estimate": 0.004017403346735516
+      "required_token_positions_per_second": 19062.560320617198,
+      "required_pages_per_second_upper_estimate": 0.003970228601262888
     },
     {
       "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": 28933.59601700338,
-      "required_pages_per_second_upper_estimate": 0.0017345249021762843
+      "required_token_positions_per_second": 28593.840480925795,
+      "required_pages_per_second_upper_estimate": 0.0017141570765650985
     },
     {
       "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": 28933.59601700338,
-      "required_pages_per_second_upper_estimate": 0.00224945982488296
+      "required_token_positions_per_second": 28593.840480925795,
+      "required_pages_per_second_upper_estimate": 0.0022230453263795955
     },
     {
       "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": 28933.59601700338,
-      "required_pages_per_second_upper_estimate": 0.37112910323114606
+      "required_token_positions_per_second": 28593.840480925795,
+      "required_pages_per_second_upper_estimate": 0.36677108401541536
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13601132416406791,
-    "producer_p95_seconds": 0.13472515400007978,
-    "gpu_update_seconds": 0.955567361338429,
-    "producer_batches_per_second": 7.352328978090962,
-    "gpu_batches_per_second": 1.0464986985316616,
-    "producer_headroom": 7.025645601286449,
+    "producer_batch_seconds": 0.3603367549140639,
+    "producer_p95_seconds": 0.13810809300002802,
+    "gpu_update_seconds": 0.966921530475882,
+    "producer_batches_per_second": 2.7751817885979695,
+    "gpu_batches_per_second": 1.0342100868390407,
+    "producer_headroom": 2.683383022379944,
     "sustainable": true,
-    "buffer_seconds": 60.200743764321025,
+    "buffer_seconds": 60.91605641998057,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 123,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 62,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 697669.9205518219,
-    "upload_bytes_per_second": 5137863.883133299,
+    "generation_bytes_per_second": 436701.6601405146,
+    "upload_bytes_per_second": 5230765.082252548,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.85186201293942,
+    "estimated_full_shards_per_hour": 93.70601037179546,
     "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": [
@@ -208,6 +208,12 @@
       "bytes": 7382400,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.1139478459999737,
+      "bytes": 6306186,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790911926322536473-f9bff804/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 483900, "loss": 0.5737193822860718, "lr": 0.0003, "grad_norm": 0.14458679732612062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70826284762, "tokens_per_second": 150471.95092740306, "data_wait_s": 1.079300000128569e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 484000, "loss": 0.5864266157150269, "lr": 0.0003, "grad_norm": 0.15354699969075963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70840892372, "tokens_per_second": 152606.40277778948, "data_wait_s": 8.565000143789803e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 484100, "loss": 0.5748189091682434, "lr": 0.0003, "grad_norm": 0.15505583909122223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70855537479, "tokens_per_second": 152773.91011407136, "data_wait_s": 7.912999990367098e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 484200, "loss": 0.6031655073165894, "lr": 0.0003, "grad_norm": 0.16529147406275682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70870343274, "tokens_per_second": 148905.85149239417, "data_wait_s": 2.171400001316215e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790911926.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.444.0.1790911926322536473-f9bff804

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c03e73cc4b0875fa4eff4df5ea727bcdb75fb19084794c0b29a08693deb98ba3
-size 19227
+oid sha256:d024111c165d2faf5707d26f79b0707f4b93ba417f89e5e4c67beb05c8207203
+size 21345