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": 75878.1269031078,
-      "required_pages_per_second_upper_estimate": 0.006576923728606944
+      "required_token_positions_per_second": 75792.22007958466,
+      "required_pages_per_second_upper_estimate": 0.006569477542872812
     },
     {
       "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": 9484.765862888475,
-      "required_pages_per_second_upper_estimate": 0.0037836149125931367
+      "required_token_positions_per_second": 9474.027509948082,
+      "required_pages_per_second_upper_estimate": 0.00377933122305253
     },
     {
       "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": 18969.53172577695,
-      "required_pages_per_second_upper_estimate": 0.004388267312402597
+      "required_token_positions_per_second": 18948.055019896165,
+      "required_pages_per_second_upper_estimate": 0.004383299054474192
     },
     {
       "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": 28454.29758866542,
-      "required_pages_per_second_upper_estimate": 0.0017327174349878733
+      "required_token_positions_per_second": 28422.082529844243,
+      "required_pages_per_second_upper_estimate": 0.0017307557069250149
     },
     {
       "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": 28454.29758866542,
-      "required_pages_per_second_upper_estimate": 0.002190490594912472
+      "required_token_positions_per_second": 28422.082529844243,
+      "required_pages_per_second_upper_estimate": 0.0021880105905074275
     },
     {
       "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": 28454.29758866542,
-      "required_pages_per_second_upper_estimate": 0.3533071456431879
+      "required_token_positions_per_second": 28422.082529844243,
+      "required_pages_per_second_upper_estimate": 0.35290714242949506
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.2321696263436479,
-    "producer_p95_seconds": 0.3145763580032508,
-    "gpu_update_seconds": 0.971663416179825,
-    "producer_batches_per_second": 4.307195629973756,
-    "gpu_batches_per_second": 1.0291629625530028,
-    "producer_headroom": 4.185144419975113,
+    "producer_batch_seconds": 0.07409869129679691,
+    "producer_p95_seconds": 0.28063203200144926,
+    "gpu_update_seconds": 0.9727647497669663,
+    "producer_batches_per_second": 13.495515001669501,
+    "gpu_batches_per_second": 1.027997776687075,
+    "producer_headroom": 13.127961273575371,
     "sustainable": true,
-    "buffer_seconds": 60.24313180314915,
+    "buffer_seconds": 60.31141448555191,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 673235.7348340243,
-    "upload_bytes_per_second": 2750244.5803203126,
+    "generation_bytes_per_second": 674101.23717318,
+    "upload_bytes_per_second": 2690134.479232922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.23035828478598,
+    "estimated_full_shards_per_hour": 72.32321661184572,
     "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": 1.291220240000257,
-      "bytes": 10903036,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.2618899920016702,
       "bytes": 13332712,
@@ -562,6 +556,12 @@
       "bytes": 13241805,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.0903980459988816,
+      "bytes": 954449,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -192,3 +192,4 @@
 {"step": 299400, "loss": 0.6056878976523876, "lr": 0.0003, "grad_norm": 0.14099047461920827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43824464906, "tokens_per_second": 150362.90423187314, "data_wait_s": 1.2349002645350993e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 299500, "loss": 0.5782710295170546, "lr": 0.0003, "grad_norm": 0.14260404235478644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43839179231, "tokens_per_second": 150031.1144806239, "data_wait_s": 9.687999408924952e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 299600, "loss": 0.6311618834733963, "lr": 0.0003, "grad_norm": 0.13785370084036394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43853846715, "tokens_per_second": 150056.79425517103, "data_wait_s": 1.0927997209364548e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 299700, "loss": 0.6409095413982868, "lr": 0.0003, "grad_norm": 0.16603829426009303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43868594238, "tokens_per_second": 150535.26290135452, "data_wait_s": 1.3698998373001814e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7dcb7c9dbdb13db0e2a7e2f870a7909d0bc6f83ff37ee27c11c23e9bce284dd3
-size 413406
+oid sha256:80e80235819ce84bb7439f1ed0271074e179e6267d37bd64f2418041ab93f033
+size 415524