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": 77184.05931191861,
-      "required_pages_per_second_upper_estimate": 0.005339772314890153
+      "required_token_positions_per_second": 76824.31222316915,
+      "required_pages_per_second_upper_estimate": 0.005314884176562217
     },
     {
       "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": 9648.007413989826,
-      "required_pages_per_second_upper_estimate": 0.004039419230932443
+      "required_token_positions_per_second": 9603.039027896144,
+      "required_pages_per_second_upper_estimate": 0.0040205919067217025
     },
     {
       "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": 19296.014827979652,
-      "required_pages_per_second_upper_estimate": 0.004018851018537113
+      "required_token_positions_per_second": 19206.078055792288,
+      "required_pages_per_second_upper_estimate": 0.004000119560682664
     },
     {
       "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": 28944.02224196948,
-      "required_pages_per_second_upper_estimate": 0.0017351499384430783
+      "required_token_positions_per_second": 28809.11708368843,
+      "required_pages_per_second_upper_estimate": 0.0017270625801923789
     },
     {
       "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": 28944.02224196948,
-      "required_pages_per_second_upper_estimate": 0.0022502704180139576
+      "required_token_positions_per_second": 28809.11708368843,
+      "required_pages_per_second_upper_estimate": 0.0022397821353426863
     },
     {
       "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": 28944.02224196948,
-      "required_pages_per_second_upper_estimate": 0.37126283965020307
+      "required_token_positions_per_second": 28809.11708368843,
+      "required_pages_per_second_upper_estimate": 0.36953242112964724
     }
   ],
   "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": 0.15004442347651548,
-    "producer_p95_seconds": 0.3563740880053956,
-    "gpu_update_seconds": 0.9552231465573496,
-    "producer_batches_per_second": 6.66469287448405,
-    "gpu_batches_per_second": 1.0468758044693822,
-    "producer_headroom": 6.366268898403002,
+    "producer_batch_seconds": 0.32068818019507717,
+    "producer_p95_seconds": 0.35735172199929366,
+    "gpu_update_seconds": 0.95969619338505,
+    "producer_batches_per_second": 3.1182939121475948,
+    "gpu_batches_per_second": 1.0419964222977587,
+    "producer_headroom": 2.992614797343822,
     "sustainable": true,
-    "buffer_seconds": 60.179058233113025,
+    "buffer_seconds": 60.46086018325815,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389218.46105592756,
-    "upload_bytes_per_second": 4490590.116837372,
+    "generation_bytes_per_second": 374645.1970113286,
+    "upload_bytes_per_second": 4571887.529956868,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.51722120054598,
+    "estimated_full_shards_per_hour": 80.39013798479932,
     "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": 2.737723589001689,
-      "bytes": 9866131,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.0393312469968805,
       "bytes": 2566382,
@@ -562,6 +556,12 @@
       "bytes": 6670770,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.436859118999564,
+      "bytes": 11704717,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -422,3 +422,4 @@
 {"step": 802200, "loss": 0.5232254266738892, "lr": 3e-05, "grad_norm": 0.2604683849349294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117418832960, "tokens_per_second": 150630.56361364116, "data_wait_s": 1.561799581395462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802300, "loss": 0.5241293907165527, "lr": 3e-05, "grad_norm": 0.2602383318531195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117433410377, "tokens_per_second": 151007.73073732303, "data_wait_s": 1.0351002856623381e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802400, "loss": 0.5335376858711243, "lr": 3e-05, "grad_norm": 0.26877125234043847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117448067975, "tokens_per_second": 151791.96007460743, "data_wait_s": 1.4217002899385989e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 802500, "loss": 0.5279629230499268, "lr": 3e-05, "grad_norm": 0.2662826119642307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117462572395, "tokens_per_second": 150028.95271998906, "data_wait_s": 1.4810997527092695e-05, "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:b7255fc5bc38efb62d6d79e228e5973bb3d71c4cb9693d448addc50d4583f866
-size 904461
+oid sha256:6b05e511a3705be71f0edb39a4f2433ea3d1671f9df2209cd1c55d820e7de51e
+size 906579