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": 75778.13848370338,
-      "required_pages_per_second_upper_estimate": 0.006568256985831302
+      "required_token_positions_per_second": 75138.42897087075,
+      "required_pages_per_second_upper_estimate": 0.006512808586587906
     },
     {
       "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": 9472.267310462923,
-      "required_pages_per_second_upper_estimate": 0.0037786290531605722
+      "required_token_positions_per_second": 9392.303621358844,
+      "required_pages_per_second_upper_estimate": 0.0037467303420132617
     },
     {
       "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": 18944.534620925846,
-      "required_pages_per_second_upper_estimate": 0.004382484672129315
+      "required_token_positions_per_second": 18784.607242717688,
+      "required_pages_per_second_upper_estimate": 0.004345488287806582
     },
     {
       "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": 28416.80193138877,
-      "required_pages_per_second_upper_estimate": 0.0017304341461842356
+      "required_token_positions_per_second": 28176.910864076533,
+      "required_pages_per_second_upper_estimate": 0.0017158260387960798
     },
     {
       "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": 28416.80193138877,
-      "required_pages_per_second_upper_estimate": 0.0021876040754206932
+      "required_token_positions_per_second": 28176.910864076533,
+      "required_pages_per_second_upper_estimate": 0.0021691365969980212
     },
     {
       "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": 28416.80193138877,
-      "required_pages_per_second_upper_estimate": 0.35284157506970426
+      "required_token_positions_per_second": 28176.910864076533,
+      "required_pages_per_second_upper_estimate": 0.34986293087744186
     }
   ],
   "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": 0.10974184799985665,
-    "producer_p95_seconds": 0.3580315680010244,
-    "gpu_update_seconds": 0.9729455153593634,
-    "producer_batches_per_second": 9.112294154198189,
-    "gpu_batches_per_second": 1.0278067828193276,
-    "producer_headroom": 8.86576573196247,
+    "producer_batch_seconds": 0.14048460122634765,
+    "producer_p95_seconds": 0.3839940959987871,
+    "gpu_update_seconds": 0.9812289265268304,
+    "producer_batches_per_second": 7.118217877764469,
+    "gpu_batches_per_second": 1.0191301672481385,
+    "producer_headroom": 6.9846012869829215,
     "sustainable": true,
-    "buffer_seconds": 60.32262195228053,
+    "buffer_seconds": 60.83619344466348,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269304.3515615402,
+    "generation_bytes_per_second": 244737.26460691995,
     "upload_bytes_per_second": 3709010.270578586,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.78644476065306,
+    "estimated_full_shards_per_hour": 52.51491979270648,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -306,3 +306,4 @@
 {"step": 225600, "loss": 0.6106600761413574, "lr": 0.0003, "grad_norm": 0.14562051416478752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33021571679, "tokens_per_second": 148417.47592848324, "data_wait_s": 1.4642002497566864e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 225700, "loss": 0.614284684881568, "lr": 0.0003, "grad_norm": 0.17948723749782694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33036120078, "tokens_per_second": 147861.11340621437, "data_wait_s": 1.2522999895736575e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 225800, "loss": 0.6041085869073868, "lr": 0.0003, "grad_norm": 0.1383085496420488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33050598731, "tokens_per_second": 146707.44813275008, "data_wait_s": 1.2013999366899952e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 225900, "loss": 0.5965267233550549, "lr": 0.0003, "grad_norm": 0.14059977196096476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33065426058, "tokens_per_second": 151548.53034548528, "data_wait_s": 1.4435001503443345e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:120f5a09b3307937e147aad8684a00f2237e8abf6df714531f050adf0afb71c1
-size 657207
+oid sha256:9ea7db136d569e87cd7f22a4ed6f88d7eab7b1025f12541cf3eca3a08c375e71
+size 659325