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": 75559.51652648897,
-      "required_pages_per_second_upper_estimate": 0.00654930738867224
+      "required_token_positions_per_second": 75410.67657494516,
+      "required_pages_per_second_upper_estimate": 0.006536406318903832
     },
     {
       "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": 9444.93956581112,
-      "required_pages_per_second_upper_estimate": 0.003767727607232775
+      "required_token_positions_per_second": 9426.334571868145,
+      "required_pages_per_second_upper_estimate": 0.003760305796979474
     },
     {
       "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": 18889.87913162224,
-      "required_pages_per_second_upper_estimate": 0.004369841086582684
+      "required_token_positions_per_second": 18852.66914373629,
+      "required_pages_per_second_upper_estimate": 0.004361233210758689
     },
     {
       "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": 28334.818697433362,
-      "required_pages_per_second_upper_estimate": 0.0017254417973691374
+      "required_token_positions_per_second": 28279.003715604435,
+      "required_pages_per_second_upper_estimate": 0.0017220429578143335
     },
     {
       "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": 28334.818697433362,
-      "required_pages_per_second_upper_estimate": 0.002181292779126689
+      "required_token_positions_per_second": 28279.003715604435,
+      "required_pages_per_second_upper_estimate": 0.00217699598731974
     },
     {
       "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": 28334.818697433362,
-      "required_pages_per_second_upper_estimate": 0.3518236176842118
+      "required_token_positions_per_second": 28279.003715604435,
+      "required_pages_per_second_upper_estimate": 0.3511305824106241
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10008276722663823,
-    "producer_p95_seconds": 0.3219244930005516,
-    "gpu_update_seconds": 0.9757606108312394,
-    "producer_batches_per_second": 9.991730122084775,
-    "gpu_batches_per_second": 1.0248415327486025,
-    "producer_headroom": 9.749536687186334,
+    "producer_batch_seconds": 0.1038304983361229,
+    "producer_p95_seconds": 0.2951793419997557,
+    "gpu_update_seconds": 0.9776864941229791,
+    "producer_batches_per_second": 9.63108158031538,
+    "gpu_batches_per_second": 1.0228227617044428,
+    "producer_headroom": 9.416178384870946,
     "sustainable": true,
-    "buffer_seconds": 60.49715787153684,
+    "buffer_seconds": 60.61656263562471,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 250614.6171034554,
-    "upload_bytes_per_second": 5181173.629463062,
+    "generation_bytes_per_second": 251983.95489308212,
+    "upload_bytes_per_second": 5069648.043071617,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.77606282069918,
+    "estimated_full_shards_per_hour": 54.06989083380077,
     "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": 0.8453881149980589,
-      "bytes": 832969,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.4130192459997488,
       "bytes": 7383726,
@@ -562,6 +556,12 @@
       "bytes": 10640144,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 3.6133480419994157,
+      "bytes": 5871337,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 379500, "loss": 0.5751869510859251, "lr": 0.0003, "grad_norm": 0.15076568877499885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55547540000, "tokens_per_second": 148638.04692342275, "data_wait_s": 1.4183999155648053e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 379600, "loss": 0.5885546300560236, "lr": 0.0003, "grad_norm": 0.16276431622782345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55562066745, "tokens_per_second": 147603.81056144106, "data_wait_s": 1.1120999261038378e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 379700, "loss": 0.5921695046126842, "lr": 0.0003, "grad_norm": 0.15509385439786771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55576775665, "tokens_per_second": 149887.22972496826, "data_wait_s": 9.9280005088076e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 379800, "loss": 0.5833548456430435, "lr": 0.0003, "grad_norm": 0.1605502428370705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55591490906, "tokens_per_second": 150128.03480669996, "data_wait_s": 8.499999239575118e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d6ff3541557bb81a2e5b82f43727fedf6cf5db806a5df0c14c495e9644c036e
-size 526443
+oid sha256:68d56331f40d063e0d34ab8478bbd116f7cfacb1dec4d8b61eb1ddfafeec7086
+size 528561