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": 75451.13994058831,
-      "required_pages_per_second_upper_estimate": 0.006539913580884318
+      "required_token_positions_per_second": 75731.57497692003,
+      "required_pages_per_second_upper_estimate": 0.006564220979077449
     },
     {
       "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": 9431.39249257354,
-      "required_pages_per_second_upper_estimate": 0.003762323477171509
+      "required_token_positions_per_second": 9466.446872115004,
+      "required_pages_per_second_upper_estimate": 0.0037763071932802794
     },
     {
       "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": 18862.78498514708,
-      "required_pages_per_second_upper_estimate": 0.004363573332636347
+      "required_token_positions_per_second": 18932.89374423001,
+      "required_pages_per_second_upper_estimate": 0.004379791760178165
     },
     {
       "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": 28294.17747772062,
-      "required_pages_per_second_upper_estimate": 0.001722966960316632
+      "required_token_positions_per_second": 28399.340616345013,
+      "required_pages_per_second_upper_estimate": 0.0017293708437105106
     },
     {
       "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": 28294.17747772062,
-      "required_pages_per_second_upper_estimate": 0.002178164105531106
+      "required_token_positions_per_second": 28399.340616345013,
+      "required_pages_per_second_upper_estimate": 0.002186259855052611
     },
     {
       "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": 28294.17747772062,
-      "required_pages_per_second_upper_estimate": 0.35131898975279213
+      "required_token_positions_per_second": 28399.340616345013,
+      "required_pages_per_second_upper_estimate": 0.3526247639761229
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.0996374236796953,
+    "producer_batch_seconds": 0.07338277021877104,
     "producer_p95_seconds": 0.19691285000044445,
-    "gpu_update_seconds": 0.9771621748598477,
-    "producer_batches_per_second": 10.036389572001609,
-    "gpu_batches_per_second": 1.0233715812254276,
-    "producer_headroom": 9.807180261917788,
+    "gpu_update_seconds": 0.9735437302402512,
+    "producer_batches_per_second": 13.627177020147485,
+    "gpu_batches_per_second": 1.0271752248388677,
+    "producer_headroom": 13.266652748838613,
     "sustainable": true,
-    "buffer_seconds": 60.584054841310554,
+    "buffer_seconds": 60.359711274895574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 482475.7310035794,
+    "generation_bytes_per_second": 465588.8652590904,
     "upload_bytes_per_second": 4020639.445295186,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.52806041317497,
+    "estimated_full_shards_per_hour": 99.90453213052305,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 323500, "loss": 0.5915633793920279, "lr": 0.0003, "grad_norm": 0.15062822252435012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47351181869, "tokens_per_second": 149537.47764343666, "data_wait_s": 1.164499917649664e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323600, "loss": 0.5999933667480946, "lr": 0.0003, "grad_norm": 0.17164955116625333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47365910119, "tokens_per_second": 150341.68643233628, "data_wait_s": 1.4494999049929902e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323700, "loss": 0.6073876842856407, "lr": 0.0003, "grad_norm": 0.1726146494891655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47380602911, "tokens_per_second": 150101.78471536125, "data_wait_s": 1.389600038237404e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 323800, "loss": 0.6049603596329689, "lr": 0.0003, "grad_norm": 0.14880438064026513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47395190148, "tokens_per_second": 148905.275910344, "data_wait_s": 1.760400118655525e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4338c18f7c3faa80aceebe41b27ee88bba162d04b1d88343a79d45d68f19766f
-size 185214
+oid sha256:cb1de250ad0a51df4f4bbd055eab7eb53c32502485fb740e09754aca162cd0fe
+size 187332