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": 75886.12137867458,
-      "required_pages_per_second_upper_estimate": 0.00657761666948726
+      "required_token_positions_per_second": 75678.17757356033,
+      "required_pages_per_second_upper_estimate": 0.0065595926274887115
     },
     {
       "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": 9485.765172334322,
-      "required_pages_per_second_upper_estimate": 0.003784013552072093
+      "required_token_positions_per_second": 9459.772196695041,
+      "required_pages_per_second_upper_estimate": 0.0037736445654599654
     },
     {
       "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": 18971.530344668645,
-      "required_pages_per_second_upper_estimate": 0.0043887296577087
+      "required_token_positions_per_second": 18919.544393390082,
+      "required_pages_per_second_upper_estimate": 0.004376703622801899
     },
     {
       "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": 28457.295517002967,
-      "required_pages_per_second_upper_estimate": 0.0017328999930947156
+      "required_token_positions_per_second": 28379.316590085124,
+      "required_pages_per_second_upper_estimate": 0.0017281514855692298
     },
     {
       "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": 28457.295517002967,
-      "required_pages_per_second_upper_estimate": 0.0021907213837346934
+      "required_token_positions_per_second": 28379.316590085124,
+      "required_pages_per_second_upper_estimate": 0.002184718350081072
     },
     {
       "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": 28457.295517002967,
-      "required_pages_per_second_upper_estimate": 0.35334436987971946
+      "required_token_positions_per_second": 28379.316590085124,
+      "required_pages_per_second_upper_estimate": 0.35237613258607997
     }
   ],
   "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": 0.08455898489870606,
-    "producer_p95_seconds": 0.26589009600138525,
-    "gpu_update_seconds": 0.971561053069171,
-    "producer_batches_per_second": 11.826064388047097,
-    "gpu_batches_per_second": 1.0292713945675263,
-    "producer_headroom": 11.48974357051486,
+    "producer_batch_seconds": 0.08469954136717206,
+    "producer_p95_seconds": 0.23589178000111133,
+    "gpu_update_seconds": 0.9742306483045957,
+    "producer_batches_per_second": 11.806439372144949,
+    "gpu_batches_per_second": 1.0264509762038891,
+    "producer_headroom": 11.502195083693678,
     "sustainable": true,
-    "buffer_seconds": 60.2367852902886,
+    "buffer_seconds": 60.40230019488494,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399628.62705903745,
-    "upload_bytes_per_second": 5641073.396379055,
+    "generation_bytes_per_second": 419811.67038434587,
+    "upload_bytes_per_second": 5549334.781183126,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.75100048855154,
+    "estimated_full_shards_per_hour": 90.08181174896032,
     "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": 1.1256958809990465,
-      "bytes": 12103363,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.0281697059999715,
       "bytes": 1191797,
@@ -562,6 +556,12 @@
       "bytes": 7473998,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2958799169973645,
+      "bytes": 6217387,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -186,3 +186,4 @@
 {"step": 333600, "loss": 0.5937071666121483, "lr": 0.0003, "grad_norm": 0.1530153455090805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48831221563, "tokens_per_second": 148852.37736342993, "data_wait_s": 1.4661000022897497e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333700, "loss": 0.593877125531435, "lr": 0.0003, "grad_norm": 0.1513579364277908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48845895110, "tokens_per_second": 149776.44402365363, "data_wait_s": 1.2620999768842012e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333800, "loss": 0.558652114123106, "lr": 0.0003, "grad_norm": 0.15084433480961412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48860349082, "tokens_per_second": 146877.6915620611, "data_wait_s": 1.4217999705579132e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 333900, "loss": 0.5932654645293951, "lr": 0.0003, "grad_norm": 0.16861895595353657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48874902633, "tokens_per_second": 147974.48455094354, "data_wait_s": 1.5769997844472528e-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 sha256:22f40fcf5edbb00f2fc9626fead950e0d27f86d08a590488c9133d37c7024a2f
-size 400698
+oid sha256:3243095782372434c2f3a25fb5574b6068c4a65676e5664ae29e4ced1de4ac7f
+size 402816