Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77171.78807757741,
-      "required_pages_per_second_upper_estimate": 0.005338923362425242
+      "required_token_positions_per_second": 77066.20665014503,
+      "required_pages_per_second_upper_estimate": 0.005331618994292807
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9646.473509697176,
-      "required_pages_per_second_upper_estimate": 0.004038777017236674
+      "required_token_positions_per_second": 9633.275831268129,
+      "required_pages_per_second_upper_estimate": 0.004033251424877297
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19292.947019394353,
-      "required_pages_per_second_upper_estimate": 0.004018212074912348
+      "required_token_positions_per_second": 19266.551662536258,
+      "required_pages_per_second_upper_estimate": 0.004012714618171178
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28939.420529091527,
-      "required_pages_per_second_upper_estimate": 0.0017348740727824536
+      "required_token_positions_per_second": 28899.82749380439,
+      "required_pages_per_second_upper_estimate": 0.0017325005307720542
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28939.420529091527,
-      "required_pages_per_second_upper_estimate": 0.0022499126550784925
+      "required_token_positions_per_second": 28899.82749380439,
+      "required_pages_per_second_upper_estimate": 0.002246834470736275
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28939.420529091527,
-      "required_pages_per_second_upper_estimate": 0.371203813818339
+      "required_token_positions_per_second": 28899.82749380439,
+      "required_pages_per_second_upper_estimate": 0.37069595687336476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5477698810702663,
-    "producer_p95_seconds": 0.24567394299992884,
-    "gpu_update_seconds": 0.9553750384257584,
-    "producer_batches_per_second": 1.8255841267616593,
-    "gpu_batches_per_second": 1.0467093652015165,
-    "producer_headroom": 1.744117505254375,
+    "producer_batch_seconds": 0.08774381044534607,
+    "producer_p95_seconds": 0.22138394500143477,
+    "gpu_update_seconds": 0.956683911207679,
+    "producer_batches_per_second": 11.396815284456796,
+    "gpu_batches_per_second": 1.0452773254414203,
+    "producer_headroom": 10.903149821645584,
     "sustainable": true,
-    "buffer_seconds": 60.18862742082278,
+    "buffer_seconds": 60.27108640608378,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384209.61872743437,
-    "upload_bytes_per_second": 5199632.347178384,
+    "generation_bytes_per_second": 384811.1413895828,
+    "upload_bytes_per_second": 5341250.6831181925,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.44244023673318,
+    "estimated_full_shards_per_hour": 82.57151299729931,
     "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": 1.0941710249999232,
-      "bytes": 2736868,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.062443419000374,
       "bytes": 5796244,
@@ -562,6 +556,12 @@
       "bytes": 1090330,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.3477415189991007,
+      "bytes": 15713674,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 898400, "loss": 0.5231099128723145, "lr": 3e-05, "grad_norm": 0.32097007019166945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131500370868, "tokens_per_second": 153466.88469262508, "data_wait_s": 1.2566999430418946e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 898500, "loss": 0.5456814765930176, "lr": 3e-05, "grad_norm": 0.35816702376225146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131515103268, "tokens_per_second": 153075.12642993315, "data_wait_s": 1.594099921931047e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 898600, "loss": 0.5198956727981567, "lr": 3e-05, "grad_norm": 0.3374929245476675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131529786288, "tokens_per_second": 153065.37951583124, "data_wait_s": 1.7922999177244492e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 898700, "loss": 0.48246288299560547, "lr": 3e-05, "grad_norm": 0.3180735630012253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131544319783, "tokens_per_second": 151395.8981096793, "data_wait_s": 1.941499976965133e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3e3168e857e788a3f04b32f782abc3483877be740cb4b80da2d5009c3d2ee770
-size 289779
+oid sha256:83b0597ba262933ee6e594a5295c5c2f27e1acb7cb4ad9410360658d9e212bbe
+size 291897