Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75948.7080288513,
-      "required_pages_per_second_upper_estimate": 0.006583041521700163
+      "required_token_positions_per_second": 75338.2944670028,
+      "required_pages_per_second_upper_estimate": 0.0065301324212381625
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9493.588503606412,
-      "required_pages_per_second_upper_estimate": 0.0037871343958857553
+      "required_token_positions_per_second": 9417.28680837535,
+      "required_pages_per_second_upper_estimate": 0.0037566965088460785
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18987.177007212824,
-      "required_pages_per_second_upper_estimate": 0.0043923492377164405
+      "required_token_positions_per_second": 18834.5736167507,
+      "required_pages_per_second_upper_estimate": 0.00435704712906097
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28480.765510819238,
-      "required_pages_per_second_upper_estimate": 0.0017343291925805855
+      "required_token_positions_per_second": 28251.860425126044,
+      "required_pages_per_second_upper_estimate": 0.0017203900738340387
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28480.765510819238,
-      "required_pages_per_second_upper_estimate": 0.002192528168827771
+      "required_token_positions_per_second": 28251.860425126044,
+      "required_pages_per_second_upper_estimate": 0.002174906421681276
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28480.765510819238,
-      "required_pages_per_second_upper_estimate": 0.35363578865390116
+      "required_token_positions_per_second": 28251.860425126044,
+      "required_pages_per_second_upper_estimate": 0.35079355358563197
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08471556918750878,
-    "producer_p95_seconds": 0.2687213740009611,
-    "gpu_update_seconds": 0.9707604238901905,
-    "producer_batches_per_second": 11.804205644733472,
-    "gpu_batches_per_second": 1.0301202803392375,
-    "producer_headroom": 11.459055675368443,
+    "producer_batch_seconds": 0.222423910140634,
+    "producer_p95_seconds": 0.21826453300127469,
+    "gpu_update_seconds": 0.9786258173430237,
+    "producer_batches_per_second": 4.495919523075198,
+    "gpu_batches_per_second": 1.0218410165337835,
+    "producer_headroom": 4.399822917977923,
     "sustainable": true,
-    "buffer_seconds": 60.18714628119181,
+    "buffer_seconds": 60.67480067526747,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421993.5805979263,
+    "generation_bytes_per_second": 430745.4716338979,
     "upload_bytes_per_second": 6375454.56635416,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.54999888852446,
+    "estimated_full_shards_per_hour": 92.4279509712477,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 54900, "loss": 0.6329132784157991, "lr": 0.0003, "grad_norm": 0.1287076918518521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8035088021, "tokens_per_second": 149553.46658519574, "data_wait_s": 1.0747000487754121e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 55000, "loss": 0.536142909899354, "lr": 0.0003, "grad_norm": 0.15281186596519372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8049776223, "tokens_per_second": 150058.34292751199, "data_wait_s": 9.393001164426096e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 55100, "loss": 0.6263116523623466, "lr": 0.0003, "grad_norm": 0.1413291128281125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8064578298, "tokens_per_second": 136813.92486464858, "data_wait_s": 8.421000529779121e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 55200, "loss": 0.6435331888496876, "lr": 0.0003, "grad_norm": 0.1433544753403895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8079270498, "tokens_per_second": 150078.95223917294, "data_wait_s": 9.573999705025926e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dfae68843919ff7e1b9ff82d4563d64b3ffc664b773ca643062c5cefd066b68d
-size 322332
+oid sha256:5b73f4562aa08a1125478826dad8a7bf0ba64ed5152f787bcd2674cf89028e43
+size 324450