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": 77032.11897381807,
-      "required_pages_per_second_upper_estimate": 0.005329260729750201
+      "required_token_positions_per_second": 77183.19318778208,
+      "required_pages_per_second_upper_estimate": 0.005339712394412703
     },
     {
       "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": 9629.014871727259,
-      "required_pages_per_second_upper_estimate": 0.004031467450096488
+      "required_token_positions_per_second": 9647.89914847276,
+      "required_pages_per_second_upper_estimate": 0.0040393739024212885
     },
     {
       "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": 19258.029743454517,
-      "required_pages_per_second_upper_estimate": 0.004010939727164571
+      "required_token_positions_per_second": 19295.79829694552,
+      "required_pages_per_second_upper_estimate": 0.004018805920833011
     },
     {
       "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": 28887.044615181778,
-      "required_pages_per_second_upper_estimate": 0.001731734216716954
+      "required_token_positions_per_second": 28943.69744541828,
+      "required_pages_per_second_upper_estimate": 0.0017351304673857693
     },
     {
       "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": 28887.044615181778,
-      "required_pages_per_second_upper_estimate": 0.002245840658149308
+      "required_token_positions_per_second": 28943.69744541828,
+      "required_pages_per_second_upper_estimate": 0.0022502451665107307
     },
     {
       "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": 28887.044615181778,
-      "required_pages_per_second_upper_estimate": 0.3705319918315796
+      "required_token_positions_per_second": 28943.69744541828,
+      "required_pages_per_second_upper_estimate": 0.3712586735087836
     }
   ],
   "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": 0.627025644757623,
-    "producer_p95_seconds": 0.2556276089999301,
-    "gpu_update_seconds": 0.9571072558066189,
-    "producer_batches_per_second": 1.5948311019823604,
-    "gpu_batches_per_second": 1.044814981741239,
-    "producer_headroom": 1.5264244194933831,
+    "producer_batch_seconds": 0.11612920839056073,
+    "producer_p95_seconds": 0.2421711389979464,
+    "gpu_update_seconds": 0.9552338657539629,
+    "producer_batches_per_second": 8.611098050688877,
+    "gpu_batches_per_second": 1.0468640569089365,
+    "producer_headroom": 8.225612479345951,
     "sustainable": true,
-    "buffer_seconds": 60.29775711581699,
+    "buffer_seconds": 60.17973354249966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299137.4796276202,
-    "upload_bytes_per_second": 5707589.346256835,
+    "generation_bytes_per_second": 300560.0494674516,
+    "upload_bytes_per_second": 5720458.859196847,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.18793956395582,
+    "estimated_full_shards_per_hour": 64.49318993585263,
     "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": 1.1592712370002118,
-      "bytes": 7445713,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0245760489997338,
       "bytes": 5205900,
@@ -562,6 +556,12 @@
       "bytes": 5869308,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.611063919001026,
+      "bytes": 11001266,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 707000, "loss": 0.5401789546012878, "lr": 8e-05, "grad_norm": 0.22340848497574609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103485085267, "tokens_per_second": 153346.46700654848, "data_wait_s": 1.2869000784121454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707100, "loss": 0.5317685008049011, "lr": 8e-05, "grad_norm": 0.21699219140543216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103499593639, "tokens_per_second": 149133.01091071288, "data_wait_s": 1.6828002117108554e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707200, "loss": 0.5109959244728088, "lr": 8e-05, "grad_norm": 0.20483624901390615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103514080865, "tokens_per_second": 150895.4630753194, "data_wait_s": 1.7610000213608146e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 707300, "loss": 0.5474433898925781, "lr": 8e-05, "grad_norm": 0.21330609410834672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103528584357, "tokens_per_second": 151187.28304752088, "data_wait_s": 1.585899735800922e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:689d06b6dcb9f77419b9447213dc4b09ac3b7e1928a5ee4959f888a633d659d4
-size 473493
+oid sha256:db53777273a723ac852725aaa001976b8b1a86365c0295f754f790d4137cf3c7
+size 475611