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": 76947.63837552589,
-      "required_pages_per_second_upper_estimate": 0.005323416165938353
+      "required_token_positions_per_second": 77200.38596675586,
+      "required_pages_per_second_upper_estimate": 0.005340901830754852
     },
     {
       "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": 9618.454796940736,
-      "required_pages_per_second_upper_estimate": 0.004027046167302809
+      "required_token_positions_per_second": 9650.048245844482,
+      "required_pages_per_second_upper_estimate": 0.004040273684612572
     },
     {
       "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": 19236.90959388147,
-      "required_pages_per_second_upper_estimate": 0.004006540956984305
+      "required_token_positions_per_second": 19300.096491688964,
+      "required_pages_per_second_upper_estimate": 0.004019701121447053
     },
     {
       "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": 28855.364390822207,
-      "required_pages_per_second_upper_estimate": 0.0017298350356394997
+      "required_token_positions_per_second": 28950.144737533446,
+      "required_pages_per_second_upper_estimate": 0.0017355169727036285
     },
     {
       "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": 28855.364390822207,
-      "required_pages_per_second_upper_estimate": 0.0022433776600519315
+      "required_token_positions_per_second": 28950.144737533446,
+      "required_pages_per_second_upper_estimate": 0.002250746415114044
     },
     {
       "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": 28855.364390822207,
-      "required_pages_per_second_upper_estimate": 0.3701256319290698
+      "required_token_positions_per_second": 28950.144737533446,
+      "required_pages_per_second_upper_estimate": 0.37134137244947407
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1856973111486866,
-    "producer_p95_seconds": 0.24561469699983718,
-    "gpu_update_seconds": 0.9581580612024354,
-    "producer_batches_per_second": 5.385107591564999,
-    "gpu_batches_per_second": 1.0436691402930487,
-    "producer_headroom": 5.159784249300436,
+    "producer_batch_seconds": 0.0645229489452106,
+    "producer_p95_seconds": 0.21727616199859767,
+    "gpu_update_seconds": 0.9550211320413458,
+    "producer_batches_per_second": 15.498361689096788,
+    "gpu_batches_per_second": 1.0470972488980557,
+    "producer_headroom": 14.801262925107439,
     "sustainable": true,
-    "buffer_seconds": 60.36395785575343,
+    "buffer_seconds": 60.16633131860478,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416958.6725018514,
+    "generation_bytes_per_second": 416706.89679242903,
     "upload_bytes_per_second": 5001816.133671999,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.46962481776863,
+    "estimated_full_shards_per_hour": 89.41559961156514,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 418000, "loss": 0.5443013906478882, "lr": 0.0003, "grad_norm": 0.13174783431493042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61184254096, "tokens_per_second": 149585.14254806645, "data_wait_s": 1.0854000720428303e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418100, "loss": 0.6076033115386963, "lr": 0.0003, "grad_norm": 0.1634113193470032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61198785933, "tokens_per_second": 150698.53566064817, "data_wait_s": 1.0841999028343707e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418200, "loss": 0.5855352282524109, "lr": 0.0003, "grad_norm": 0.1395451328877831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61213494863, "tokens_per_second": 152492.80641446833, "data_wait_s": 1.1574000382097438e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 418300, "loss": 0.5724180936813354, "lr": 0.0003, "grad_norm": 0.15005547982988734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61228169417, "tokens_per_second": 152253.36935442977, "data_wait_s": 1.5689001884311438e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:851b905adcc7c2b861d04cd43eceeb6ec96d7fee37a33ac78a9d797dcf097aad
-size 450195
+oid sha256:8bc7860f994cae990ade9e79f26577683a69d4ddcfc10b45822eb3a13acdc92c
+size 452313