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": 76017.0494337998,
-      "required_pages_per_second_upper_estimate": 0.006588965181471391
+      "required_token_positions_per_second": 76184.22602069385,
+      "required_pages_per_second_upper_estimate": 0.0066034556243181825
     },
     {
       "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": 9502.131179224974,
-      "required_pages_per_second_upper_estimate": 0.003790542196914383
+      "required_token_positions_per_second": 9523.028252586731,
+      "required_pages_per_second_upper_estimate": 0.0037988783519174773
     },
     {
       "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": 19004.26235844995,
-      "required_pages_per_second_upper_estimate": 0.004396301633033237
+      "required_token_positions_per_second": 19046.056505173463,
+      "required_pages_per_second_upper_estimate": 0.004405969973325862
     },
     {
       "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": 28506.39353767492,
-      "required_pages_per_second_upper_estimate": 0.0017358898049562315
+      "required_token_positions_per_second": 28569.084757760193,
+      "required_pages_per_second_upper_estimate": 0.0017397073713440125
     },
     {
       "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": 28506.39353767492,
-      "required_pages_per_second_upper_estimate": 0.0021945010852780404
+      "required_token_positions_per_second": 28569.084757760193,
+      "required_pages_per_second_upper_estimate": 0.0021993272289406087
     },
     {
       "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": 28506.39353767492,
-      "required_pages_per_second_upper_estimate": 0.35395400297596036
+      "required_token_positions_per_second": 28569.084757760193,
+      "required_pages_per_second_upper_estimate": 0.3547324181154028
     }
   ],
   "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": 0.15557834879695065,
-    "producer_p95_seconds": 0.40123429899540497,
-    "gpu_update_seconds": 0.9698876837387219,
-    "producer_batches_per_second": 6.427629600987256,
-    "gpu_batches_per_second": 1.0310472199679876,
-    "producer_headroom": 6.234078785631975,
+    "producer_batch_seconds": 0.27603550144544897,
+    "producer_p95_seconds": 0.38482849999854807,
+    "gpu_update_seconds": 0.9677593886689002,
+    "producer_batches_per_second": 3.622722420715957,
+    "gpu_batches_per_second": 1.0333146975463032,
+    "producer_headroom": 3.505923635189193,
     "sustainable": true,
-    "buffer_seconds": 60.13303639180076,
+    "buffer_seconds": 60.001082097471816,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 603209.4473612143,
+    "generation_bytes_per_second": 553790.4601112283,
     "upload_bytes_per_second": 5599528.092696146,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.71735270322476,
+    "estimated_full_shards_per_hour": 118.83054115774762,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 311100, "loss": 0.5741534139961004, "lr": 0.0003, "grad_norm": 0.13278086682221193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45537060365, "tokens_per_second": 150512.8172377953, "data_wait_s": 7.806003850419074e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 311200, "loss": 0.6199047211557627, "lr": 0.0003, "grad_norm": 0.17804002496439653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45551524664, "tokens_per_second": 148861.33297630894, "data_wait_s": 1.0080999345518649e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 311300, "loss": 0.5691689476370811, "lr": 0.0003, "grad_norm": 0.14703916731569824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45566015724, "tokens_per_second": 148500.84986881696, "data_wait_s": 7.180999091360718e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 311400, "loss": 0.5803543459624052, "lr": 0.0003, "grad_norm": 0.1543395265847266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45580506588, "tokens_per_second": 149135.81556980906, "data_wait_s": 1.5265999536495656e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ce91760df8dff90a20b88a8e4af8810806f36f44540238a86a88f9ec60c8038
-size 663561
+oid sha256:683df71a2b4d6cf4cc5756279086e2095babfbbc8e285c1ee98e6b30d696d2f4
+size 665679