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": 74817.06559769281,
-      "required_pages_per_second_upper_estimate": 0.006484953623888865
+      "required_token_positions_per_second": 74768.47626238667,
+      "required_pages_per_second_upper_estimate": 0.006480742023453073
     },
     {
       "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": 9352.133199711601,
-      "required_pages_per_second_upper_estimate": 0.003730705760216851
+      "required_token_positions_per_second": 9346.059532798334,
+      "required_pages_per_second_upper_estimate": 0.0037282828836757353
     },
     {
       "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": 18704.266399423203,
-      "required_pages_per_second_upper_estimate": 0.004326902847660952
+      "required_token_positions_per_second": 18692.119065596667,
+      "required_pages_per_second_upper_estimate": 0.004324092775765957
     },
     {
       "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": 28056.399599134802,
-      "required_pages_per_second_upper_estimate": 0.0017084875350348709
+      "required_token_positions_per_second": 28038.178598395,
+      "required_pages_per_second_upper_estimate": 0.0017073779716880195
     },
     {
       "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": 28056.399599134802,
-      "required_pages_per_second_upper_estimate": 0.002159859306226274
+      "required_token_positions_per_second": 28038.178598395,
+      "required_pages_per_second_upper_estimate": 0.0021584566031503656
     },
     {
       "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": 28056.399599134802,
-      "required_pages_per_second_upper_estimate": 0.34836658429212414
+      "required_token_positions_per_second": 28038.178598395,
+      "required_pages_per_second_upper_estimate": 0.34814034044470243
     }
   ],
   "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": 0.1271477255705804,
-    "producer_p95_seconds": 0.3719332789987675,
-    "gpu_update_seconds": 0.9854436205297206,
-    "producer_batches_per_second": 7.864867385652875,
-    "gpu_batches_per_second": 1.0147713975381512,
-    "producer_headroom": 7.750383391503888,
+    "producer_batch_seconds": 0.4937962972733203,
+    "producer_p95_seconds": 0.40013938899937784,
+    "gpu_update_seconds": 0.9860840247869262,
+    "producer_batches_per_second": 2.0251265664037406,
+    "gpu_batches_per_second": 1.0141123624998192,
+    "producer_headroom": 1.996944955302329,
     "sustainable": true,
-    "buffer_seconds": 60.11206085231296,
+    "buffer_seconds": 60.1511255120025,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273285.97819557064,
+    "generation_bytes_per_second": 255505.77086627402,
     "upload_bytes_per_second": 5595221.703859815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.64080914879169,
+    "estimated_full_shards_per_hour": 54.82559055796781,
     "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

@@ -405,3 +405,4 @@
 {"step": 235500, "loss": 0.5931294355541468, "lr": 0.0003, "grad_norm": 0.16225231457587036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34470705573, "tokens_per_second": 147665.02847612434, "data_wait_s": 1.0807998478412628e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 235600, "loss": 0.5948690362274647, "lr": 0.0003, "grad_norm": 0.1593855203694616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34485172640, "tokens_per_second": 130516.4335955507, "data_wait_s": 8.487004379276186e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 235700, "loss": 0.5529085770249367, "lr": 0.0003, "grad_norm": 0.14916057235719044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34499840230, "tokens_per_second": 148132.54662396782, "data_wait_s": 1.955799962161109e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 235800, "loss": 0.5849977470934391, "lr": 0.0003, "grad_norm": 0.1671746701560514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34514496775, "tokens_per_second": 148640.04987322216, "data_wait_s": 9.06200148165226e-06, "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 sha256:2ce3d1c00fc6e95ce373e33b8292d1cff149afbf48dcf79aaa30bd701998aa70
-size 868455
+oid sha256:7abe667994470b7e0072e676ccea999633e2c892bd4dfa11e518c742e8923eb8
+size 870573