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": 77042.82497483374,
-      "required_pages_per_second_upper_estimate": 0.00533000139574181
+      "required_token_positions_per_second": 76859.77989241168,
+      "required_pages_per_second_upper_estimate": 0.005317337912216746
     },
     {
       "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": 9630.353121854218,
-      "required_pages_per_second_upper_estimate": 0.004032027747478805
+      "required_token_positions_per_second": 9607.47248655146,
+      "required_pages_per_second_upper_estimate": 0.004022448103279538
     },
     {
       "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": 19260.706243708435,
-      "required_pages_per_second_upper_estimate": 0.004011497171583403
+      "required_token_positions_per_second": 19214.94497310292,
+      "required_pages_per_second_upper_estimate": 0.004001966305722134
     },
     {
       "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": 28891.05936556265,
-      "required_pages_per_second_upper_estimate": 0.0017319748948721185
+      "required_token_positions_per_second": 28822.417459654378,
+      "required_pages_per_second_upper_estimate": 0.0017278599174230392
     },
     {
       "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": 28891.05936556265,
-      "required_pages_per_second_upper_estimate": 0.0022461527873324
+      "required_token_positions_per_second": 28822.417459654378,
+      "required_pages_per_second_upper_estimate": 0.0022408161810718676
     },
     {
       "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": 28891.05936556265,
-      "required_pages_per_second_upper_estimate": 0.37058348873876235
+      "required_token_positions_per_second": 28822.417459654378,
+      "required_pages_per_second_upper_estimate": 0.36970302407170735
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07681934602332774,
-    "producer_p95_seconds": 0.17553877499994996,
-    "gpu_update_seconds": 0.9569742545666448,
-    "producer_batches_per_second": 13.017554193917896,
-    "gpu_batches_per_second": 1.0449601911734177,
-    "producer_headroom": 12.457464221005479,
+    "producer_batch_seconds": 0.4567267198124938,
+    "producer_p95_seconds": 0.184863974000109,
+    "gpu_update_seconds": 0.9592533325388709,
+    "producer_batches_per_second": 2.189493096463775,
+    "gpu_batches_per_second": 1.0424774833497676,
+    "producer_headroom": 2.1002785493537277,
     "sustainable": true,
-    "buffer_seconds": 60.28937803769862,
+    "buffer_seconds": 60.43295994994887,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 270075.930806423,
-    "upload_bytes_per_second": 4571425.307156558,
+    "generation_bytes_per_second": 259488.05878523772,
+    "upload_bytes_per_second": 4561519.041951923,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.95200770515936,
+    "estimated_full_shards_per_hour": 55.68009684245919,
     "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": 1.1746671439996135,
-      "bytes": 6226498,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0685757479996028,
       "bytes": 7863934,
@@ -562,6 +556,12 @@
       "bytes": 2526198,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2877745800014964,
+      "bytes": 5791906,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -95,3 +95,4 @@
 {"step": 449700, "loss": 0.5876931548118591, "lr": 0.0003, "grad_norm": 0.17943939720067453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65821948113, "tokens_per_second": 150426.37173868914, "data_wait_s": 1.3969000065117143e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 449800, "loss": 0.5955415368080139, "lr": 0.0003, "grad_norm": 0.1597044843710986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65836597883, "tokens_per_second": 152744.1483349692, "data_wait_s": 1.5227000403683633e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 449900, "loss": 0.5834906697273254, "lr": 0.0003, "grad_norm": 0.16927764105128051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65851154163, "tokens_per_second": 152264.63207802546, "data_wait_s": 1.3408998711383902e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 450000, "loss": 0.6337220668792725, "lr": 0.0003, "grad_norm": 0.19062176168096287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65865880260, "tokens_per_second": 152074.55981661563, "data_wait_s": 1.1875999916810542e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:386c7687ad68ad146acb54234a71b1a39009d1a50937f608f9401092c485200e
-size 206394
+oid sha256:fd78b0bdfb87c16927fa35644d8ff37319d83cf7fd16ceb74d3af8dd9da930db
+size 208512