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": 76772.17526222316,
-      "required_pages_per_second_upper_estimate": 0.005311277220629556
+      "required_token_positions_per_second": 76660.01445962541,
+      "required_pages_per_second_upper_estimate": 0.005303517676056926
     },
     {
       "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": 9596.521907777895,
-      "required_pages_per_second_upper_estimate": 0.004017863324621135
+      "required_token_positions_per_second": 9582.501807453176,
+      "required_pages_per_second_upper_estimate": 0.0040119934013881624
     },
     {
       "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": 19193.04381555579,
-      "required_pages_per_second_upper_estimate": 0.003997404872177432
+      "required_token_positions_per_second": 19165.003614906353,
+      "required_pages_per_second_upper_estimate": 0.003991564837851972
     },
     {
       "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": 28789.565723333686,
-      "required_pages_per_second_upper_estimate": 0.0017258905059923107
+      "required_token_positions_per_second": 28747.505422359533,
+      "required_pages_per_second_upper_estimate": 0.0017233690551712712
     },
     {
       "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": 28789.565723333686,
-      "required_pages_per_second_upper_estimate": 0.0022382621030724507
+      "required_token_positions_per_second": 28747.505422359533,
+      "required_pages_per_second_upper_estimate": 0.0022349920996754223
     },
     {
       "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": 28789.565723333686,
-      "required_pages_per_second_upper_estimate": 0.36928163727163177
+      "required_token_positions_per_second": 28747.505422359533,
+      "required_pages_per_second_upper_estimate": 0.368742132891568
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11201718858637832,
-    "producer_p95_seconds": 0.32497797900578007,
-    "gpu_update_seconds": 0.9603479352796053,
-    "producer_batches_per_second": 8.9272013752504,
-    "gpu_batches_per_second": 1.04128926950715,
-    "producer_headroom": 8.573219408546974,
+    "producer_batch_seconds": 0.11892233741423297,
+    "producer_p95_seconds": 0.31795231599971885,
+    "gpu_update_seconds": 0.9617530145240238,
+    "producer_batches_per_second": 8.40884918463028,
+    "gpu_batches_per_second": 1.0397679912601103,
+    "producer_headroom": 8.087236051996053,
     "sustainable": true,
-    "buffer_seconds": 60.50191992261513,
+    "buffer_seconds": 60.590439915013505,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263476.0285722965,
+    "generation_bytes_per_second": 254351.59812368272,
     "upload_bytes_per_second": 4843450.842176247,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.53582232357665,
+    "estimated_full_shards_per_hour": 54.57793195517408,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 758900, "loss": 0.5560118556022644, "lr": 5e-05, "grad_norm": 0.2524935765127414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111081192519, "tokens_per_second": 137861.00600475605, "data_wait_s": 1.4269993698690087e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 759000, "loss": 0.5519813895225525, "lr": 5e-05, "grad_norm": 0.2668603073262277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111095670229, "tokens_per_second": 149642.5407720595, "data_wait_s": 1.2448006600607187e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 759100, "loss": 0.5515175461769104, "lr": 5e-05, "grad_norm": 0.24843087476895098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111110173626, "tokens_per_second": 149278.94468124522, "data_wait_s": 1.0984003893099725e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 759200, "loss": 0.5073992609977722, "lr": 5e-05, "grad_norm": 0.23980937587591555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111124782897, "tokens_per_second": 151043.34325093732, "data_wait_s": 1.4990997442509979e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fab43c9fc05ad631f816002228271ec2cd41f9cb914c3ba90dd17605979c15e1
-size 833784
+oid sha256:bf723f7d74eadd93b47eb8668fa8461517d43246064877a13c9451ff39b0b35a
+size 835902