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": 76184.22602069385,
-      "required_pages_per_second_upper_estimate": 0.0066034556243181825
+      "required_token_positions_per_second": 75588.87378964707,
+      "required_pages_per_second_upper_estimate": 0.006551852001837482
     },
     {
       "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": 9523.028252586731,
-      "required_pages_per_second_upper_estimate": 0.0037988783519174773
+      "required_token_positions_per_second": 9448.609223705884,
+      "required_pages_per_second_upper_estimate": 0.003769191488633271
     },
     {
       "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": 19046.056505173463,
-      "required_pages_per_second_upper_estimate": 0.004405969973325862
+      "required_token_positions_per_second": 18897.218447411768,
+      "required_pages_per_second_upper_estimate": 0.004371538908123045
     },
     {
       "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": 28569.084757760193,
-      "required_pages_per_second_upper_estimate": 0.0017397073713440125
+      "required_token_positions_per_second": 28345.827671117655,
+      "required_pages_per_second_upper_estimate": 0.0017261121861068898
     },
     {
       "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": 28569.084757760193,
-      "required_pages_per_second_upper_estimate": 0.0021993272289406087
+      "required_token_positions_per_second": 28345.827671117655,
+      "required_pages_per_second_upper_estimate": 0.0021821402803956958
     },
     {
       "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": 28569.084757760193,
-      "required_pages_per_second_upper_estimate": 0.3547324181154028
+      "required_token_positions_per_second": 28345.827671117655,
+      "required_pages_per_second_upper_estimate": 0.3519603122927059
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "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,
+    "producer_batch_seconds": 0.11436324779703,
+    "producer_p95_seconds": 0.38095969099958893,
+    "gpu_update_seconds": 0.9753816441977212,
+    "producer_batches_per_second": 8.744067865008377,
+    "gpu_batches_per_second": 1.0252397161139197,
+    "producer_headroom": 8.528803291148328,
     "sustainable": true,
-    "buffer_seconds": 60.001082097471816,
+    "buffer_seconds": 60.473661940258715,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 553790.4601112283,
-    "upload_bytes_per_second": 5599528.092696146,
+    "generation_bytes_per_second": 556483.2526261441,
+    "upload_bytes_per_second": 5659244.709862863,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.83054115774762,
+    "estimated_full_shards_per_hour": 119.40835174644701,
     "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": 1.4521962930011796,
-      "bytes": 2626843,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2158071900012146,
       "bytes": 5739703,
@@ -562,6 +556,12 @@
       "bytes": 7275318,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.0238104009986273,
+      "bytes": 4591810,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -310,3 +310,4 @@
 {"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}
+{"step": 311500, "loss": 0.5991575866937637, "lr": 0.0003, "grad_norm": 0.15397654657214074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45595202813, "tokens_per_second": 150432.2112460396, "data_wait_s": 1.3062999641988426e-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 sha256:683df71a2b4d6cf4cc5756279086e2095babfbbc8e285c1ee98e6b30d696d2f4
-size 665679
+oid sha256:12b40affc7ae29b8c3b95846dfb959a3fb56ae6e04071c1839356ea0f75a84a5
+size 667797