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": 77068.39121065705,
-      "required_pages_per_second_upper_estimate": 0.005331770127257907
+      "required_token_positions_per_second": 77156.8734634909,
+      "required_pages_per_second_upper_estimate": 0.0053378915348160656
     },
     {
       "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": 9633.548901332131,
-      "required_pages_per_second_upper_estimate": 0.004033365753610744
+      "required_token_positions_per_second": 9644.609182936363,
+      "required_pages_per_second_upper_estimate": 0.0040379964625534915
     },
     {
       "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": 19267.097802664262,
-      "required_pages_per_second_upper_estimate": 0.004012828364757158
+      "required_token_positions_per_second": 19289.218365872726,
+      "required_pages_per_second_upper_estimate": 0.004017435494715
     },
     {
       "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": 28900.646703996394,
-      "required_pages_per_second_upper_estimate": 0.0017325496411722033
+      "required_token_positions_per_second": 28933.827548809088,
+      "required_pages_per_second_upper_estimate": 0.0017345387821545077
     },
     {
       "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": 28900.646703996394,
-      "required_pages_per_second_upper_estimate": 0.0022468981607253785
+      "required_token_positions_per_second": 28933.827548809088,
+      "required_pages_per_second_upper_estimate": 0.0022494778254624533
     },
     {
       "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": 28900.646703996394,
-      "required_pages_per_second_upper_estimate": 0.3707064648221084
+      "required_token_positions_per_second": 28933.827548809088,
+      "required_pages_per_second_upper_estimate": 0.3711320730725502
     }
   ],
   "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": 0.05618902426559913,
-    "producer_p95_seconds": 0.14967549799985136,
-    "gpu_update_seconds": 0.9566567932951072,
-    "producer_batches_per_second": 17.79707003405351,
-    "gpu_batches_per_second": 1.0453069554396843,
-    "producer_headroom": 17.02568794882608,
+    "producer_batch_seconds": 0.04515752917969351,
+    "producer_p95_seconds": 0.1476149949999126,
+    "gpu_update_seconds": 0.9555597147788346,
+    "producer_batches_per_second": 22.144701407837015,
+    "gpu_batches_per_second": 1.0465070728012547,
+    "producer_headroom": 21.160584561135195,
     "sustainable": true,
-    "buffer_seconds": 60.26937797759175,
+    "buffer_seconds": 60.20026203106658,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353716.7327595308,
-    "upload_bytes_per_second": 34790.06957199078,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 369188.6947630823,
+    "upload_bytes_per_second": 431426.06842612405,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.89937674607698,
+    "estimated_full_shards_per_hour": 79.2192996231971,
     "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": [
@@ -184,6 +184,12 @@
       "bytes": 49312,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.2620071970000026,
+      "bytes": 1106661,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -32,3 +32,4 @@
 {"step": 648200, "loss": 0.5496170520782471, "lr": 0.0001, "grad_norm": 0.17056241102373523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94879908023, "tokens_per_second": 154547.97673710226, "data_wait_s": 1.6008999864425277e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648300, "loss": 0.5251896977424622, "lr": 0.0001, "grad_norm": 0.17190561853882502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94894560854, "tokens_per_second": 152289.052323069, "data_wait_s": 1.3718999980483204e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648400, "loss": 0.6239676475524902, "lr": 0.0001, "grad_norm": 0.21783978781609148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94909152120, "tokens_per_second": 152245.14727116865, "data_wait_s": 1.3509999917005189e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 648500, "loss": 0.5002675652503967, "lr": 0.0001, "grad_norm": 0.17678204568389247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94923748418, "tokens_per_second": 150892.88864365392, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:44310fec6b38e6d87ff05a4a8072fed57c57b19b5e90550ee9ab82716f87a7e4
-size 72177
+oid sha256:fc245dfbac7faaf0b464c807916f4ea840f227c9b85d9244dd2453c30c90ed80
+size 74295