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": 75695.91670086801,
-      "required_pages_per_second_upper_estimate": 0.0065611302100843346
+      "required_token_positions_per_second": 75123.9163420815,
+      "required_pages_per_second_upper_estimate": 0.0065115506687063755
     },
     {
       "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": 9461.989587608501,
-      "required_pages_per_second_upper_estimate": 0.003774529115848293
+      "required_token_positions_per_second": 9390.489542760188,
+      "required_pages_per_second_upper_estimate": 0.0037460066789373656
     },
     {
       "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": 18923.979175217002,
-      "required_pages_per_second_upper_estimate": 0.004377729531528062
+      "required_token_positions_per_second": 18780.979085520376,
+      "required_pages_per_second_upper_estimate": 0.00434464897749237
     },
     {
       "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": 28385.968762825505,
-      "required_pages_per_second_upper_estimate": 0.001728556567987866
+      "required_token_positions_per_second": 28171.468628280563,
+      "required_pages_per_second_upper_estimate": 0.0017154946351893097
     },
     {
       "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": 28385.968762825505,
-      "required_pages_per_second_upper_estimate": 0.0021852304527530198
+      "required_token_positions_per_second": 28171.468628280563,
+      "required_pages_per_second_upper_estimate": 0.0021687176386213742
     },
     {
       "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": 28385.968762825505,
-      "required_pages_per_second_upper_estimate": 0.3524587303081255
+      "required_token_positions_per_second": 28171.468628280563,
+      "required_pages_per_second_upper_estimate": 0.34979535652284743
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10955438575786047,
-    "producer_p95_seconds": 0.3829696110042278,
-    "gpu_update_seconds": 0.9740023400648579,
-    "producer_batches_per_second": 9.127886511181963,
-    "gpu_batches_per_second": 1.0266915785165476,
-    "producer_headroom": 8.890582821737684,
+    "producer_batch_seconds": 0.24495871825774884,
+    "producer_p95_seconds": 0.37026148100267164,
+    "gpu_update_seconds": 0.9814184828207689,
+    "producer_batches_per_second": 4.082320511441387,
+    "gpu_batches_per_second": 1.0189333271224164,
+    "producer_headroom": 4.006464802726912,
     "sustainable": true,
-    "buffer_seconds": 60.38814508402119,
+    "buffer_seconds": 60.847945934887676,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491021.5619626832,
+    "generation_bytes_per_second": 593617.4727335217,
     "upload_bytes_per_second": 4664112.175691686,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.36179680023548,
+    "estimated_full_shards_per_hour": 63.68824547054405,
     "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": [

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -331,3 +331,4 @@
 {"step": 73100, "loss": 0.6249210871756077, "lr": 0.0003, "grad_norm": 0.13517386636139803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10698270266, "tokens_per_second": 149889.29134254056, "data_wait_s": 8.770999556872994e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 73200, "loss": 0.6195316538214684, "lr": 0.0003, "grad_norm": 0.15515854729561093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10712902577, "tokens_per_second": 149556.26445771017, "data_wait_s": 7.280999852810055e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 73300, "loss": 0.6037278231233358, "lr": 0.0003, "grad_norm": 0.13094053326599112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10727361084, "tokens_per_second": 146893.6334221823, "data_wait_s": 1.1030999303329736e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 73400, "loss": 0.6100993007421494, "lr": 0.0003, "grad_norm": 0.12337211974140161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10741768291, "tokens_per_second": 146735.5201714556, "data_wait_s": 1.044199598254636e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c546b801db4774f082755b1174af2004c8597e37234d6b49fa2bec0d5c01fbe
-size 710157
+oid sha256:86f53cdac3e6fe84fc14275093f21404bc4579e88c31bc67f0ea8a9d095b4dba
+size 712275