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": 74630.09735383728,
-      "required_pages_per_second_upper_estimate": 0.006468747690378149
+      "required_token_positions_per_second": 74512.20476791683,
+      "required_pages_per_second_upper_estimate": 0.006458529059826593
     },
     {
       "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": 9328.76216922966,
-      "required_pages_per_second_upper_estimate": 0.0037213827067295595
+      "required_token_positions_per_second": 9314.025595989604,
+      "required_pages_per_second_upper_estimate": 0.003715504067332697
     },
     {
       "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": 18657.52433845932,
-      "required_pages_per_second_upper_estimate": 0.004316089894489135
+      "required_token_positions_per_second": 18628.051191979208,
+      "required_pages_per_second_upper_estimate": 0.004309271800760623
     },
     {
       "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": 27986.28650768898,
-      "required_pages_per_second_upper_estimate": 0.0017042180156207818
+      "required_token_positions_per_second": 27942.07678796881,
+      "required_pages_per_second_upper_estimate": 0.0017015258756402968
     },
     {
       "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": 27986.28650768898,
-      "required_pages_per_second_upper_estimate": 0.0021544618063613157
+      "required_token_positions_per_second": 27942.07678796881,
+      "required_pages_per_second_upper_estimate": 0.00215105841975692
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27986.28650768898,
-      "required_pages_per_second_upper_estimate": 0.3474960143497893
+      "required_token_positions_per_second": 27942.07678796881,
+      "required_pages_per_second_upper_estimate": 0.3469470775912787
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4804735872499606,
-    "producer_p95_seconds": 0.2357324789991253,
-    "gpu_update_seconds": 0.9879124189057367,
-    "producer_batches_per_second": 2.0812798591564663,
-    "gpu_batches_per_second": 1.0122354784320378,
-    "producer_headroom": 2.056122220079055,
+    "producer_batch_seconds": 0.07868971369539679,
+    "producer_p95_seconds": 0.22994795799968415,
+    "gpu_update_seconds": 0.9894754856555461,
+    "producer_batches_per_second": 12.708141293675824,
+    "gpu_batches_per_second": 1.0106364578981775,
+    "producer_headroom": 12.574394278339186,
     "sustainable": true,
-    "buffer_seconds": 60.26265755324994,
+    "buffer_seconds": 60.35800462498831,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318906.81582723005,
+    "generation_bytes_per_second": 314673.62711454113,
     "upload_bytes_per_second": 5316085.367041503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.42997890579869,
+    "estimated_full_shards_per_hour": 67.52163515164543,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 205000, "loss": 0.613105034455657, "lr": 0.0003, "grad_norm": 0.14022459818757185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30004511655, "tokens_per_second": 146295.43586384007, "data_wait_s": 2.1854999431525357e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205100, "loss": 0.5696049407124519, "lr": 0.0003, "grad_norm": 0.24330457566497732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30019051201, "tokens_per_second": 114408.83670990342, "data_wait_s": 1.8001999706029892e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205200, "loss": 0.5707568153738976, "lr": 0.0003, "grad_norm": 0.15176623283134827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30033541888, "tokens_per_second": 146343.10543456714, "data_wait_s": 1.715399957902264e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 205300, "loss": 0.584257559850812, "lr": 0.0003, "grad_norm": 0.14663389437398983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30048220354, "tokens_per_second": 148249.03032386256, "data_wait_s": 1.8435999663779512e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:760b3eb6fd30d54bf12456d023a39b1e42e6da99a5342f1a4cc92f203d3f91b6
-size 217767
+oid sha256:97b4b86dacc4226ca5da52c1559b593dbfd1dbfdb1b8ac7205aae9084a7495e0
+size 219885