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": 75329.03647818745,
-      "required_pages_per_second_upper_estimate": 0.006529329962231807
+      "required_token_positions_per_second": 75594.51657615908,
+      "required_pages_per_second_upper_estimate": 0.00655234110427083
     },
     {
       "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": 9416.129559773432,
-      "required_pages_per_second_upper_estimate": 0.003756234865076365
+      "required_token_positions_per_second": 9449.314572019885,
+      "required_pages_per_second_upper_estimate": 0.003769472862621623
     },
     {
       "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": 18832.259119546863,
-      "required_pages_per_second_upper_estimate": 0.004356511710866623
+      "required_token_positions_per_second": 18898.62914403977,
+      "required_pages_per_second_upper_estimate": 0.00437186524796052
     },
     {
       "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": 28248.388679320295,
-      "required_pages_per_second_upper_estimate": 0.001720178662729311
+      "required_token_positions_per_second": 28347.943716059654,
+      "required_pages_per_second_upper_estimate": 0.0017262410421418277
     },
     {
       "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": 28248.388679320295,
-      "required_pages_per_second_upper_estimate": 0.0021746391570787423
+      "required_token_positions_per_second": 28347.943716059654,
+      "required_pages_per_second_upper_estimate": 0.002182303179392918
     },
     {
       "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": 28248.388679320295,
-      "required_pages_per_second_upper_estimate": 0.35075044612191036
+      "required_token_positions_per_second": 28347.943716059654,
+      "required_pages_per_second_upper_estimate": 0.35198658648893866
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1402635221089099,
-    "producer_p95_seconds": 0.4073327320002136,
-    "gpu_update_seconds": 0.9787460911085587,
-    "producer_batches_per_second": 7.1294373972980205,
-    "gpu_batches_per_second": 1.0217154470240268,
-    "producer_headroom": 6.9779089844086135,
+    "producer_batch_seconds": 0.2576498914058334,
+    "producer_p95_seconds": 0.41716916199948173,
+    "gpu_update_seconds": 0.975308836398489,
+    "producer_batches_per_second": 3.8812358683468835,
+    "gpu_batches_per_second": 1.025316251304241,
+    "producer_headroom": 3.785403638545478,
     "sustainable": true,
-    "buffer_seconds": 60.68225764873064,
+    "buffer_seconds": 60.469147856706314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383310.43860726524,
-    "upload_bytes_per_second": 4021417.176453192,
+    "generation_bytes_per_second": 445589.72292000084,
+    "upload_bytes_per_second": 4022126.850080259,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.24949711478679,
+    "estimated_full_shards_per_hour": 95.61318174076098,
     "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": 1.2323054039989074,
-      "bytes": 11963716,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 0.7193277299993497,
       "bytes": 1174751,
@@ -562,6 +556,12 @@
       "bytes": 1112881,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.220746667997446,
+      "bytes": 11991108,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 350300, "loss": 0.6137314103543758, "lr": 0.0003, "grad_norm": 0.18405110382549328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51274718620, "tokens_per_second": 148061.49646377063, "data_wait_s": 1.1934993381146342e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350400, "loss": 0.5734471511095762, "lr": 0.0003, "grad_norm": 0.1576187271171644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51289254835, "tokens_per_second": 147763.8878159974, "data_wait_s": 1.3365002814680338e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350500, "loss": 0.5647007338702679, "lr": 0.0003, "grad_norm": 0.16194227915709652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51303831394, "tokens_per_second": 148074.64501590794, "data_wait_s": 1.7079997633118182e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 350600, "loss": 0.588532218709588, "lr": 0.0003, "grad_norm": 0.16141536980723187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51318537286, "tokens_per_second": 148859.55574717984, "data_wait_s": 1.3267999747768044e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c97502e1534c0b534185a36a5e16b0a4a8d85e7dc861d1c2560b04f031a470d
-size 757536
+oid sha256:27cbcfc157d4893512abc8447921959a84c8fc7b5bf0d49d21feed5d1b0b5338
+size 759654