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": 76446.73503825655,
-      "required_pages_per_second_upper_estimate": 0.005288762510810191
+      "required_token_positions_per_second": 76478.53876361436,
+      "required_pages_per_second_upper_estimate": 0.005290962766325254
     },
     {
       "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": 9555.841879782069,
-      "required_pages_per_second_upper_estimate": 0.004000831446394866
+      "required_token_positions_per_second": 9559.817345451795,
+      "required_pages_per_second_upper_estimate": 0.0040024958908536174
     },
     {
       "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": 19111.683759564137,
-      "required_pages_per_second_upper_estimate": 0.003980459718123333
+      "required_token_positions_per_second": 19119.63469090359,
+      "required_pages_per_second_upper_estimate": 0.003982115687441181
     },
     {
       "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": 28667.525639346204,
-      "required_pages_per_second_upper_estimate": 0.0017185743893016797
+      "required_token_positions_per_second": 28679.452036355382,
+      "required_pages_per_second_upper_estimate": 0.001719289358591826
     },
     {
       "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": 28667.525639346204,
-      "required_pages_per_second_upper_estimate": 0.002228774023340023
+      "required_token_positions_per_second": 28679.452036355382,
+      "required_pages_per_second_upper_estimate": 0.00222970124824881
     },
     {
       "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": 28667.525639346204,
-      "required_pages_per_second_upper_estimate": 0.36771623811067333
+      "required_token_positions_per_second": 28679.452036355382,
+      "required_pages_per_second_upper_estimate": 0.3678692171259397
     }
   ],
   "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": 0.11194435960112514,
-    "producer_p95_seconds": 0.3492206829978386,
-    "gpu_update_seconds": 0.9644362177548067,
-    "producer_batches_per_second": 8.933009251767153,
-    "gpu_batches_per_second": 1.0368752039694085,
-    "producer_headroom": 8.61531765594301,
+    "producer_batch_seconds": 0.09251984485894127,
+    "producer_p95_seconds": 0.3400648809983977,
+    "gpu_update_seconds": 0.9640351553771715,
+    "producer_batches_per_second": 10.808491967584168,
+    "gpu_batches_per_second": 1.037306569601974,
+    "producer_headroom": 10.419766233362914,
     "sustainable": true,
-    "buffer_seconds": 60.75948171855282,
+    "buffer_seconds": 60.734214788761804,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442278.0245702655,
-    "upload_bytes_per_second": 3903976.4259499377,
+    "generation_bytes_per_second": 450674.6243787286,
+    "upload_bytes_per_second": 3962868.3216520934,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.90256836729978,
+    "estimated_full_shards_per_hour": 96.70428322335619,
     "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": 1.479657814001257,
-      "bytes": 836473,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.9911568649986293,
       "bytes": 11961950,
@@ -562,6 +556,12 @@
       "bytes": 6917854,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1665768739985651,
+      "bytes": 5996687,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -370,3 +370,4 @@
 {"step": 837000, "loss": 0.52271568775177, "lr": 3e-05, "grad_norm": 0.2812109963595893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122512721369, "tokens_per_second": 149911.48262071869, "data_wait_s": 1.103399699786678e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 837100, "loss": 0.5420165061950684, "lr": 3e-05, "grad_norm": 0.2981183837032969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122527383846, "tokens_per_second": 152028.0753820676, "data_wait_s": 9.708994184620678e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 837200, "loss": 0.5118781328201294, "lr": 3e-05, "grad_norm": 0.2969276606246413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122542060561, "tokens_per_second": 152149.6545058889, "data_wait_s": 1.3486001989804208e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 837300, "loss": 0.4833088517189026, "lr": 3e-05, "grad_norm": 0.30236235154591024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122556690050, "tokens_per_second": 151832.62821011522, "data_wait_s": 9.859002602752298e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4834f12a15447dbf5c802f8e477930bf1a4348cd8431739a21b8ac851efa51ce
-size 793542
+oid sha256:caf655f9f501b3d6eead77475869c795d0599436afc0104f1376ff68beded490
+size 795660