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": 75249.05969920107,
-      "required_pages_per_second_upper_estimate": 0.0065223977777285605
+      "required_token_positions_per_second": 75597.30134555085,
+      "required_pages_per_second_upper_estimate": 0.006552582481023776
     },
     {
       "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": 9406.132462400134,
-      "required_pages_per_second_upper_estimate": 0.003752246873464231
+      "required_token_positions_per_second": 9449.662668193856,
+      "required_pages_per_second_upper_estimate": 0.0037696117233899217
     },
     {
       "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": 18812.264924800267,
-      "required_pages_per_second_upper_estimate": 0.004351886405797807
+      "required_token_positions_per_second": 18899.325336387712,
+      "required_pages_per_second_upper_estimate": 0.00437202629980765
     },
     {
       "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": 28218.3973872004,
-      "required_pages_per_second_upper_estimate": 0.0017183523503913583
+      "required_token_positions_per_second": 28348.98800458157,
+      "required_pages_per_second_upper_estimate": 0.0017263046338338538
     },
     {
       "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": 28218.3973872004,
-      "required_pages_per_second_upper_estimate": 0.0021723303443901424
+      "required_token_positions_per_second": 28348.98800458157,
+      "required_pages_per_second_upper_estimate": 0.0021823835716141106
     },
     {
       "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": 28218.3973872004,
-      "required_pages_per_second_upper_estimate": 0.35037805464817906
+      "required_token_positions_per_second": 28348.98800458157,
+      "required_pages_per_second_upper_estimate": 0.351999553057372
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.4140352259844349,
-    "producer_p95_seconds": 0.33745574000204215,
-    "gpu_update_seconds": 0.9797863294866233,
-    "producer_batches_per_second": 2.4152534307252247,
-    "gpu_batches_per_second": 1.0206306925347366,
-    "producer_headroom": 2.366432293670242,
+    "producer_batch_seconds": 0.06821863382813831,
+    "producer_p95_seconds": 0.21936518700022134,
+    "gpu_update_seconds": 0.9752729090552269,
+    "producer_batches_per_second": 14.658751485983695,
+    "gpu_batches_per_second": 1.0253540221564514,
+    "producer_headroom": 14.296283204852948,
     "sustainable": true,
-    "buffer_seconds": 60.746752428170645,
+    "buffer_seconds": 60.466920361424066,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 205198.1074427795,
-    "upload_bytes_per_second": 6119637.723563503,
+    "generation_bytes_per_second": 228746.10889763758,
+    "upload_bytes_per_second": 6106163.102498237,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.03073708975352,
+    "estimated_full_shards_per_hour": 49.083590032547434,
     "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": 0.756701359001454,
-      "bytes": 1071685,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.2869488490014191,
       "bytes": 16651954,
@@ -562,6 +556,12 @@
       "bytes": 15814135,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.1238995200001227,
+      "bytes": 2339467,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -186,3 +186,4 @@
 {"step": 18600, "loss": 0.6644454542547464, "lr": 0.0003, "grad_norm": 0.15401647829993756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2721860495, "tokens_per_second": 149481.10288003285, "data_wait_s": 1.3980999938212335e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 18700, "loss": 0.6925698667764664, "lr": 0.0003, "grad_norm": 0.15461326582955442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2736545256, "tokens_per_second": 149196.83346456004, "data_wait_s": 1.079399953596294e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 18800, "loss": 0.682854663580656, "lr": 0.0003, "grad_norm": 0.1324900247318256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2751283713, "tokens_per_second": 148798.14179879194, "data_wait_s": 1.4485001884168014e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 18900, "loss": 0.6635485030710697, "lr": 0.0003, "grad_norm": 0.12805478054388889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2765781333, "tokens_per_second": 147688.75364777775, "data_wait_s": 1.0233001376036555e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33d3abd4f8b006d8dc16b279507fda090ea0a8ce3c6fc0855fdde826e5540fcb
-size 394924
+oid sha256:153e5aefb8c81f9b589705102d8d5b767fd458db23eb1b60082ed96d4aefedf4
+size 397042