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": 75785.3298382113,
-      "required_pages_per_second_upper_estimate": 0.006568880314213726
+      "required_token_positions_per_second": 75728.53264384157,
+      "required_pages_per_second_upper_estimate": 0.00656395727735693
     },
     {
       "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": 9473.166229776412,
-      "required_pages_per_second_upper_estimate": 0.0037789876455147646
+      "required_token_positions_per_second": 9466.066580480196,
+      "required_pages_per_second_upper_estimate": 0.0037761554892612875
     },
     {
       "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": 18946.332459552825,
-      "required_pages_per_second_upper_estimate": 0.004382900570454787
+      "required_token_positions_per_second": 18932.13316096039,
+      "required_pages_per_second_upper_estimate": 0.004379615812624547
     },
     {
       "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": 28419.498689329233,
-      "required_pages_per_second_upper_estimate": 0.0017305983645940143
+      "required_token_positions_per_second": 28398.199741440585,
+      "required_pages_per_second_upper_estimate": 0.0017293013704145388
     },
     {
       "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": 28419.498689329233,
-      "required_pages_per_second_upper_estimate": 0.0021878116793120537
+      "required_token_positions_per_second": 28398.199741440585,
+      "required_pages_per_second_upper_estimate": 0.0021861720273443244
     },
     {
       "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": 28419.498689329233,
-      "required_pages_per_second_upper_estimate": 0.35287505977785655
+      "required_token_positions_per_second": 28398.199741440585,
+      "required_pages_per_second_upper_estimate": 0.3526105981280726
     }
   ],
   "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": 0.08685869073457297,
-    "producer_p95_seconds": 0.2827593040019565,
-    "gpu_update_seconds": 0.9728531914738203,
-    "producer_batches_per_second": 11.512952722898493,
-    "gpu_batches_per_second": 1.0279043218073363,
-    "producer_headroom": 11.200412799759007,
+    "producer_batch_seconds": 0.669559345390752,
+    "producer_p95_seconds": 0.2889369149997947,
+    "gpu_update_seconds": 0.9735828415790089,
+    "producer_batches_per_second": 1.4935195914805792,
+    "gpu_batches_per_second": 1.0271339605555767,
+    "producer_headroom": 1.4540650478275827,
     "sustainable": true,
-    "buffer_seconds": 60.31689787137686,
+    "buffer_seconds": 60.36213617789855,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403102.2392158579,
-    "upload_bytes_per_second": 3958681.5709599303,
+    "generation_bytes_per_second": 329757.69641906576,
+    "upload_bytes_per_second": 4005387.717017378,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.49635679585268,
+    "estimated_full_shards_per_hour": 70.75832528523425,
     "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": 1.2559754029989563,
-      "bytes": 9926404,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9813094319997617,
       "bytes": 1049873,
@@ -562,6 +556,12 @@
       "bytes": 1217353,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.4986801279992505,
+      "bytes": 15911648,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -131,3 +131,4 @@
 {"step": 293300, "loss": 0.5915417168289423, "lr": 0.0003, "grad_norm": 0.1528488919679106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42930985270, "tokens_per_second": 152309.3695231858, "data_wait_s": 1.55180023284629e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 293400, "loss": 0.6367223151028156, "lr": 0.0003, "grad_norm": 0.1769232731724887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42945773257, "tokens_per_second": 150833.97254747324, "data_wait_s": 7.929000275908038e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 293500, "loss": 0.5993944443762302, "lr": 0.0003, "grad_norm": 0.15140161998271245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42960534743, "tokens_per_second": 152025.60115668707, "data_wait_s": 9.122002666117623e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 293600, "loss": 0.6076928023248911, "lr": 0.0003, "grad_norm": 0.1773359722816775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42975015545, "tokens_per_second": 132135.99491782277, "data_wait_s": 1.7972000932786614e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ff351a13a7b1ef1b3048d66d6c3c40b355b7d648d8fca793658ed23fdd8f8b3
-size 283425
+oid sha256:824e6da6fdddc45ff247b2fc09f631a81557eade5189843a13d85ed269775c66
+size 285543