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": 76444.99336729341,
-      "required_pages_per_second_upper_estimate": 0.005288642017971727
+      "required_token_positions_per_second": 77155.43231783768,
+      "required_pages_per_second_upper_estimate": 0.0053377918330158565
     },
     {
       "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": 9555.624170911677,
-      "required_pages_per_second_upper_estimate": 0.004000740296237111
+      "required_token_positions_per_second": 9644.42903972971,
+      "required_pages_per_second_upper_estimate": 0.004037921040354684
     },
     {
       "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": 19111.248341823353,
-      "required_pages_per_second_upper_estimate": 0.003980369032090666
+      "required_token_positions_per_second": 19288.85807945942,
+      "required_pages_per_second_upper_estimate": 0.004017360456556499
     },
     {
       "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": 28666.87251273503,
-      "required_pages_per_second_upper_estimate": 0.0017185352353586063
+      "required_token_positions_per_second": 28933.28711918913,
+      "required_pages_per_second_upper_estimate": 0.0017345063842239797
     },
     {
       "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": 28666.87251273503,
-      "required_pages_per_second_upper_estimate": 0.0022287232456188045
+      "required_token_positions_per_second": 28933.28711918913,
+      "required_pages_per_second_upper_estimate": 0.0022494358094366004
     },
     {
       "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": 28666.87251273503,
-      "required_pages_per_second_upper_estimate": 0.3677078605037779
+      "required_token_positions_per_second": 28933.28711918913,
+      "required_pages_per_second_upper_estimate": 0.37112514102165356
     }
   ],
   "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": 0.07458110692186892,
-    "producer_p95_seconds": 0.2620379159998265,
-    "gpu_update_seconds": 0.9644581908165112,
-    "producer_batches_per_second": 13.408221482252856,
-    "gpu_batches_per_second": 1.0368515810451038,
-    "producer_headroom": 12.931669032840668,
+    "producer_batch_seconds": 0.25612885124996865,
+    "producer_p95_seconds": 0.27027381399966544,
+    "gpu_update_seconds": 0.955577563175091,
+    "producer_batches_per_second": 3.9042848750531864,
+    "gpu_batches_per_second": 1.0464875260123383,
+    "producer_headroom": 3.7308470268446885,
     "sustainable": true,
-    "buffer_seconds": 60.76086602144021,
+    "buffer_seconds": 60.201386480030735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389704.2455578036,
-    "upload_bytes_per_second": 4775112.971789076,
+    "generation_bytes_per_second": 374511.9241445784,
+    "upload_bytes_per_second": 4813867.800523102,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.62145924616414,
+    "estimated_full_shards_per_hour": 80.36154073002828,
     "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": 1.3114747930012527,
-      "bytes": 6594730,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1013290299997607,
       "bytes": 5841716,
@@ -562,6 +556,12 @@
       "bytes": 7910132,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.3195631069975207,
+      "bytes": 10098822,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 949200, "loss": 0.5632185339927673, "lr": 3e-05, "grad_norm": 0.35542638318195346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138936514089, "tokens_per_second": 153708.60155259547, "data_wait_s": 1.0625997674651444e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949300, "loss": 0.5321933031082153, "lr": 3e-05, "grad_norm": 0.3530456270577749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138951280039, "tokens_per_second": 153581.8055451676, "data_wait_s": 8.073999197222292e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 949400, "loss": 0.4543458819389343, "lr": 3e-05, "grad_norm": 0.3367170802597107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138966051757, "tokens_per_second": 154100.82524088165, "data_wait_s": 8.947998139774427e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 949500, "loss": 0.5028177499771118, "lr": 3e-05, "grad_norm": 0.3515104382144916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138980615292, "tokens_per_second": 152106.50922828596, "data_wait_s": 1.1082000128226355e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:505386530d0e7e5d8a189114517486723da79f53846349077ece0e182b584649
-size 520089
+oid sha256:29abce8f972ae4fe72f2f62e6c8642d748e53ec26d70e29bb8743ed7e290c8fc
+size 522207