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": 76183.06290441012,
-      "required_pages_per_second_upper_estimate": 0.0052705210607870544
+      "required_token_positions_per_second": 76918.45525580634,
+      "required_pages_per_second_upper_estimate": 0.005321397210001988
     },
     {
       "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": 9522.882863051265,
-      "required_pages_per_second_upper_estimate": 0.003987032194352214
+      "required_token_positions_per_second": 9614.806906975793,
+      "required_pages_per_second_upper_estimate": 0.004025518871951092
     },
     {
       "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": 19045.76572610253,
-      "required_pages_per_second_upper_estimate": 0.003966730730128724
+      "required_token_positions_per_second": 19229.613813951586,
+      "required_pages_per_second_upper_estimate": 0.004005021438427564
     },
     {
       "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": 28568.6485891538,
-      "required_pages_per_second_upper_estimate": 0.0017126468611191588
+      "required_token_positions_per_second": 28844.420720927377,
+      "required_pages_per_second_upper_estimate": 0.0017291789793393208
     },
     {
       "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": 28568.6485891538,
-      "required_pages_per_second_upper_estimate": 0.002221086767601744
+      "required_token_positions_per_second": 28844.420720927377,
+      "required_pages_per_second_upper_estimate": 0.002242526837853725
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28568.6485891538,
-      "required_pages_per_second_upper_estimate": 0.36644794947671017
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28844.420720927377,
+      "required_pages_per_second_upper_estimate": 0.36998525828205614
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1080172583673118,
-    "producer_p95_seconds": 0.3121216090003145,
-    "gpu_update_seconds": 0.9677741638257497,
-    "producer_batches_per_second": 9.257779868838258,
-    "gpu_batches_per_second": 1.0332989217720558,
-    "producer_headroom": 8.959440171447802,
+    "producer_batch_seconds": 0.5582834919607933,
+    "producer_p95_seconds": 0.3259290969981521,
+    "gpu_update_seconds": 0.9585215895821634,
+    "producer_batches_per_second": 1.7912046736109246,
+    "gpu_batches_per_second": 1.043273318899283,
+    "producer_headroom": 1.7169083510165435,
     "sustainable": true,
-    "buffer_seconds": 60.00199815719648,
+    "buffer_seconds": 60.386860143676294,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290043.6476237112,
-    "upload_bytes_per_second": 5558726.6317567155,
+    "generation_bytes_per_second": 271474.7153317913,
+    "upload_bytes_per_second": 5412349.792632056,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.23661490949156,
+    "estimated_full_shards_per_hour": 58.25215430226616,
     "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": 1.58055241799957,
-      "bytes": 15825794,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.061300241999561,
       "bytes": 7669220,
@@ -562,6 +556,12 @@
       "bytes": 6233393,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9377798560017254,
+      "bytes": 884750,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 471600, "loss": 0.5859622359275818, "lr": 0.0003, "grad_norm": 0.1678073753660921, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69027484151, "tokens_per_second": 152556.14185780848, "data_wait_s": 1.3237000530352816e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 471700, "loss": 0.5782145261764526, "lr": 0.0003, "grad_norm": 0.2143807544097022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69042167529, "tokens_per_second": 152489.50233741896, "data_wait_s": 9.118997695622966e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 471800, "loss": 0.5450863242149353, "lr": 0.0003, "grad_norm": 0.14126341136660386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69056806674, "tokens_per_second": 152460.0685611972, "data_wait_s": 1.6985999536700547e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 471900, "loss": 0.5733816027641296, "lr": 0.0003, "grad_norm": 0.15933333398687086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69071504283, "tokens_per_second": 131852.69168520567, "data_wait_s": 1.1574000382097438e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5417966498e5cc5899911fe8dacfffc017a452e33d9d4133fb3de9f8da028375
-size 674151
+oid sha256:b65f125fc1edc1e1d934f54c5b924896a6bf2e7500282121273724da01631707
+size 676269