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": 76637.64762129493,
-      "required_pages_per_second_upper_estimate": 0.005301970286283009
+      "required_token_positions_per_second": 76812.79570879469,
+      "required_pages_per_second_upper_estimate": 0.005314087437375796
     },
     {
       "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": 9579.705952661867,
-      "required_pages_per_second_upper_estimate": 0.004010822835371128
+      "required_token_positions_per_second": 9601.599463599336,
+      "required_pages_per_second_upper_estimate": 0.004019989191212149
     },
     {
       "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": 19159.411905323734,
-      "required_pages_per_second_upper_estimate": 0.003990400232209211
+      "required_token_positions_per_second": 19203.19892719867,
+      "required_pages_per_second_upper_estimate": 0.0039995199141243405
     },
     {
       "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": 28739.117857985602,
-      "required_pages_per_second_upper_estimate": 0.0017228662334941233
+      "required_token_positions_per_second": 28804.798390798005,
+      "required_pages_per_second_upper_estimate": 0.0017268036811478602
     },
     {
       "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": 28739.117857985602,
-      "required_pages_per_second_upper_estimate": 0.002234340003438346
+      "required_token_positions_per_second": 28804.798390798005,
+      "required_pages_per_second_upper_estimate": 0.0022394463759663782
     },
     {
       "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": 28739.117857985602,
-      "required_pages_per_second_upper_estimate": 0.3686345462216442
+      "required_token_positions_per_second": 28804.798390798005,
+      "required_pages_per_second_upper_estimate": 0.3694770255743
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.25139971454689203,
-    "producer_p95_seconds": 0.33119152700237464,
-    "gpu_update_seconds": 0.9620337039091679,
-    "producer_batches_per_second": 3.9777292579760517,
-    "gpu_batches_per_second": 1.039464621599595,
-    "producer_headroom": 3.826709611198567,
+    "producer_batch_seconds": 0.12028743826550681,
+    "producer_p95_seconds": 0.32645044200035045,
+    "gpu_update_seconds": 0.9598400802844169,
+    "producer_batches_per_second": 8.313420041357356,
+    "gpu_batches_per_second": 1.0418402195745806,
+    "producer_headroom": 7.979553759934525,
     "sustainable": true,
-    "buffer_seconds": 60.60812334627757,
+    "buffer_seconds": 60.46992505791827,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386690.2303490771,
-    "upload_bytes_per_second": 4918713.34758047,
+    "generation_bytes_per_second": 424303.09910171427,
+    "upload_bytes_per_second": 4817714.282944879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.97472174505458,
+    "estimated_full_shards_per_hour": 91.04556779659816,
     "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": 1.2929117179992318,
-      "bytes": 11923823,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.1343894960009493,
       "bytes": 6177604,
@@ -562,6 +556,12 @@
       "bytes": 1047762,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 2.737723589001689,
+      "bytes": 9866131,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -343,3 +343,4 @@
 {"step": 794300, "loss": 0.5353758335113525, "lr": 3e-05, "grad_norm": 0.2582250826480756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116263171118, "tokens_per_second": 152911.1696900838, "data_wait_s": 9.776005754247308e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 794400, "loss": 0.5692605376243591, "lr": 3e-05, "grad_norm": 0.2803299069318174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116277915773, "tokens_per_second": 153034.29447523592, "data_wait_s": 1.064100069925189e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 794500, "loss": 0.5542953014373779, "lr": 3e-05, "grad_norm": 0.2731092819199093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116292484074, "tokens_per_second": 151676.86563222043, "data_wait_s": 1.5102996258065104e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 794600, "loss": 0.508455753326416, "lr": 3e-05, "grad_norm": 0.2578542515966335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116307039693, "tokens_per_second": 151368.67870592087, "data_wait_s": 9.748000593390316e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03fa5723fb47347f02a890f85f26bfba0ce8ab9f965022dc1899a11e0f7e34e9
-size 735573
+oid sha256:86bfadad30d05946f9628f47692aa7640f1c685d200773ba35a3a939b447a731
+size 737691