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": 77291.51700734878,
-      "required_pages_per_second_upper_estimate": 0.005347206487596218
+      "required_token_positions_per_second": 77122.04513909131,
+      "required_pages_per_second_upper_estimate": 0.005335482030521255
     },
     {
       "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": 9661.439625918598,
-      "required_pages_per_second_upper_estimate": 0.004045043017570538
+      "required_token_positions_per_second": 9640.255642386413,
+      "required_pages_per_second_upper_estimate": 0.004036173726037493
     },
     {
       "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": 19322.879251837196,
-      "required_pages_per_second_upper_estimate": 0.004024446169564141
+      "required_token_positions_per_second": 19280.511284772827,
+      "required_pages_per_second_upper_estimate": 0.004015622039343061
     },
     {
       "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": 28984.318877755795,
-      "required_pages_per_second_upper_estimate": 0.0017375656602290675
+      "required_token_positions_per_second": 28920.766927159242,
+      "required_pages_per_second_upper_estimate": 0.0017337558178290142
     },
     {
       "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": 28984.318877755795,
-      "required_pages_per_second_upper_estimate": 0.0022534033042001748
+      "required_token_positions_per_second": 28920.766927159242,
+      "required_pages_per_second_upper_estimate": 0.002248462419576781
     },
     {
       "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": 28984.318877755795,
-      "required_pages_per_second_upper_estimate": 0.3717797216269134
+      "required_token_positions_per_second": 28920.766927159242,
+      "required_pages_per_second_upper_estimate": 0.37096454544142893
     }
   ],
   "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": 0.2736490696712508,
-    "producer_p95_seconds": 0.3705578119988786,
-    "gpu_update_seconds": 0.9538951084760056,
-    "producer_batches_per_second": 3.654315365301089,
-    "gpu_batches_per_second": 1.0483332927429034,
-    "producer_headroom": 3.485833551789416,
+    "producer_batch_seconds": 0.13080151705440812,
+    "producer_p95_seconds": 0.3654125960019883,
+    "gpu_update_seconds": 0.9559912456552458,
+    "producer_batches_per_second": 7.645171268036903,
+    "gpu_batches_per_second": 1.0460346834186647,
+    "producer_headroom": 7.308716803778295,
     "sustainable": true,
-    "buffer_seconds": 60.09539183398835,
+    "buffer_seconds": 60.227448476280486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 536379.5608416472,
+    "generation_bytes_per_second": 550453.4844861795,
     "upload_bytes_per_second": 5125643.458896423,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.09456747948705,
+    "estimated_full_shards_per_hour": 118.11450386942902,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -433,3 +433,4 @@
 {"step": 968300, "loss": 0.4854298233985901, "lr": 3e-05, "grad_norm": 0.3531872913909241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141732946685, "tokens_per_second": 153001.29599698883, "data_wait_s": 1.1126001481898129e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 968400, "loss": 0.5527081489562988, "lr": 3e-05, "grad_norm": 0.356889287735655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141747560366, "tokens_per_second": 151789.66124980632, "data_wait_s": 1.0071998985949904e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 968500, "loss": 0.5100990533828735, "lr": 3e-05, "grad_norm": 0.338577821665479, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141762164178, "tokens_per_second": 151992.21728515413, "data_wait_s": 1.980800152523443e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 968600, "loss": 0.5253740549087524, "lr": 3e-05, "grad_norm": 0.35511977396952704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141776659980, "tokens_per_second": 150906.8296378775, "data_wait_s": 1.0513998859096318e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57cece9185a183fc2790d3a66feea42dd969e19fe7ea647f037b19f71d8df5ad
-size 927759
+oid sha256:aadd465d4502cd4f43a06bd42b9b25f2485a77f3784d8b28e9622958ea1b41bd
+size 929877