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": 76787.71081898027,
-      "required_pages_per_second_upper_estimate": 0.005312352006493471
+      "required_token_positions_per_second": 77241.39166139596,
+      "required_pages_per_second_upper_estimate": 0.005343738699856375
     },
     {
       "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": 9598.463852372533,
-      "required_pages_per_second_upper_estimate": 0.004018676376270496
+      "required_token_positions_per_second": 9655.173957674495,
+      "required_pages_per_second_upper_estimate": 0.004042419713118764
     },
     {
       "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": 19196.927704745067,
-      "required_pages_per_second_upper_estimate": 0.003998213783870512
+      "required_token_positions_per_second": 19310.34791534899,
+      "required_pages_per_second_upper_estimate": 0.004021836222647214
     },
     {
       "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": 28795.3915571176,
-      "required_pages_per_second_upper_estimate": 0.001726239755832123
+      "required_token_positions_per_second": 28965.52187302349,
+      "required_pages_per_second_upper_estimate": 0.0017364388084967302
     },
     {
       "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": 28795.3915571176,
-      "required_pages_per_second_upper_estimate": 0.0022387150360240154
+      "required_token_positions_per_second": 28965.52187302349,
+      "required_pages_per_second_upper_estimate": 0.002251941919761524
     },
     {
       "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": 28795.3915571176,
-      "required_pages_per_second_upper_estimate": 0.36935636481211886
+      "required_token_positions_per_second": 28965.52187302349,
+      "required_pages_per_second_upper_estimate": 0.3715386138328586
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08037485001558764,
-    "producer_p95_seconds": 0.21002127400061,
-    "gpu_update_seconds": 0.9601536393474048,
-    "producer_batches_per_second": 12.44170284368883,
-    "gpu_batches_per_second": 1.0414999839813948,
-    "producer_headroom": 11.945946265046787,
+    "producer_batch_seconds": 0.057142534156326974,
+    "producer_p95_seconds": 0.20159809700089681,
+    "gpu_update_seconds": 0.9545141330855655,
+    "producer_batches_per_second": 17.50009891518396,
+    "gpu_batches_per_second": 1.0476534242268334,
+    "producer_headroom": 16.704091744938463,
     "sustainable": true,
-    "buffer_seconds": 60.489679278886506,
+    "buffer_seconds": 60.13439038439063,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 502397.54684210016,
-    "upload_bytes_per_second": 5428082.149759443,
+    "generation_bytes_per_second": 502051.06093470246,
+    "upload_bytes_per_second": 2178019.4106091666,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.80281833598379,
+    "estimated_full_shards_per_hour": 107.72847052603537,
     "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": 1.7348476629995275,
-      "bytes": 1091907,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.6013349519998883,
       "bytes": 17762321,
@@ -562,6 +556,12 @@
       "bytes": 6117925,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 122.56198037699869,
+      "bytes": 6934679,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 658500, "loss": 0.5744720697402954, "lr": 0.0001, "grad_norm": 0.17951242339152534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96386594202, "tokens_per_second": 151203.9696837603, "data_wait_s": 1.3013999705435708e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658600, "loss": 0.5846948027610779, "lr": 0.0001, "grad_norm": 0.1931559684828068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96401125533, "tokens_per_second": 150968.4826024349, "data_wait_s": 1.0240999472443946e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658700, "loss": 0.5515803694725037, "lr": 0.0001, "grad_norm": 0.1961138554767908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96415753876, "tokens_per_second": 153168.6945994227, "data_wait_s": 8.998998964671046e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 658800, "loss": 0.5704889893531799, "lr": 0.0001, "grad_norm": 0.19855481073692224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96430477585, "tokens_per_second": 153797.17199487513, "data_wait_s": 1.0458999895490706e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0eef8e0c27714085b4ad6504a8625623d756beeadb394e6997d6eda3b8a57dbb
-size 291897
+oid sha256:673fa60f374a9ab6a1f7f84903e5550cda6dab35d5bd914322631ad63d61c78f
+size 294015