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": 77051.59573317453,
-      "required_pages_per_second_upper_estimate": 0.00533060817715479
+      "required_token_positions_per_second": 77275.68885495991,
+      "required_pages_per_second_upper_estimate": 0.005346111459287582
     },
     {
       "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": 9631.449466646816,
-      "required_pages_per_second_upper_estimate": 0.004032486764149184
+      "required_token_positions_per_second": 9659.461106869989,
+      "required_pages_per_second_upper_estimate": 0.004044214652961061
     },
     {
       "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": 19262.898933293633,
-      "required_pages_per_second_upper_estimate": 0.004011953850998887
+      "required_token_positions_per_second": 19318.922213739977,
+      "required_pages_per_second_upper_estimate": 0.004023622022882602
     },
     {
       "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": 28894.34839994045,
-      "required_pages_per_second_upper_estimate": 0.0017321720674610029
+      "required_token_positions_per_second": 28978.383320609966,
+      "required_pages_per_second_upper_estimate": 0.001737209832641246
     },
     {
       "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": 28894.34839994045,
-      "required_pages_per_second_upper_estimate": 0.0022464084952883404
+      "required_token_positions_per_second": 28978.383320609966,
+      "required_pages_per_second_upper_estimate": 0.0022529418407398433
     },
     {
       "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": 28894.34839994045,
-      "required_pages_per_second_upper_estimate": 0.37062567694027077
+      "required_token_positions_per_second": 28978.383320609966,
+      "required_pages_per_second_upper_estimate": 0.37170358667295145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.45015194264055935,
-    "producer_p95_seconds": 0.3707712769974023,
-    "gpu_update_seconds": 0.9568653224952802,
-    "producer_batches_per_second": 2.2214721414597722,
-    "gpu_batches_per_second": 1.04507915219692,
-    "producer_headroom": 2.1256496570521857,
+    "producer_batch_seconds": 0.12465931710158884,
+    "producer_p95_seconds": 0.3559359540013247,
+    "gpu_update_seconds": 0.9540904920094775,
+    "producer_batches_per_second": 8.021863293098807,
+    "gpu_batches_per_second": 1.0481186096864137,
+    "producer_headroom": 7.653583496145409,
     "sustainable": true,
-    "buffer_seconds": 60.28251531720265,
+    "buffer_seconds": 60.10770099659708,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337316.2910471855,
+    "generation_bytes_per_second": 336830.0625890837,
     "upload_bytes_per_second": 5012214.322547673,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.38022373735116,
+    "estimated_full_shards_per_hour": 72.27589042906173,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 686300, "loss": 0.5422401428222656, "lr": 0.0001, "grad_norm": 0.2058293217477818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100456651841, "tokens_per_second": 152989.9583270223, "data_wait_s": 9.724004485178739e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 686400, "loss": 0.5442904829978943, "lr": 0.0001, "grad_norm": 0.19064226928362107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100471198826, "tokens_per_second": 151235.1030276125, "data_wait_s": 1.6117999621201307e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686500, "loss": 0.5686470866203308, "lr": 0.0001, "grad_norm": 0.21736728526906218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100485727691, "tokens_per_second": 150947.96057938872, "data_wait_s": 1.1701005860231817e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 686600, "loss": 0.5638381838798523, "lr": 0.0001, "grad_norm": 0.1899685754727259, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100500353538, "tokens_per_second": 151369.94095271057, "data_wait_s": 1.4338002074509859e-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 sha256:94b0dbdfa4f4b9a50238fc451edc5d199f69a2fdedc3ebb49cd435abb149a74f
-size 885399
+oid sha256:52e9cf4eedfb0b79d1e8ebb28cffebe16c691c8107eac3e0a041ea30d7b2ab10
+size 887517