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": 76423.12924518762,
-      "required_pages_per_second_upper_estimate": 0.005287129407272696
+      "required_token_positions_per_second": 76780.34560431352,
+      "required_pages_per_second_upper_estimate": 0.0053118424638531735
     },
     {
       "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": 9552.891155648453,
-      "required_pages_per_second_upper_estimate": 0.003999596039818249
+      "required_token_positions_per_second": 9597.54320053919,
+      "required_pages_per_second_upper_estimate": 0.004018290918573271
     },
     {
       "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": 19105.782311296905,
-      "required_pages_per_second_upper_estimate": 0.003979230602080925
+      "required_token_positions_per_second": 19195.08640107838,
+      "required_pages_per_second_upper_estimate": 0.003997830288875185
     },
     {
       "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": 28658.673466945354,
-      "required_pages_per_second_upper_estimate": 0.0017180437150827333
+      "required_token_positions_per_second": 28792.62960161757,
+      "required_pages_per_second_upper_estimate": 0.001726074180817158
     },
     {
       "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": 28658.673466945354,
-      "required_pages_per_second_upper_estimate": 0.002228085805872448
+      "required_token_positions_per_second": 28792.62960161757,
+      "required_pages_per_second_upper_estimate": 0.0022385003061324463
     },
     {
       "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": 28658.673466945354,
-      "required_pages_per_second_upper_estimate": 0.36760269194783746
+      "required_token_positions_per_second": 28792.62960161757,
+      "required_pages_per_second_upper_estimate": 0.36932093741252126
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0691633875702351,
-    "producer_p95_seconds": 0.17003646500052128,
-    "gpu_update_seconds": 0.9647341155510545,
-    "producer_batches_per_second": 14.458516783674087,
-    "gpu_batches_per_second": 1.0365550299097712,
-    "producer_headroom": 13.948624401477899,
+    "producer_batch_seconds": 0.0497324833905779,
+    "producer_p95_seconds": 0.16540787400026602,
+    "gpu_update_seconds": 0.9602457428357546,
+    "producer_batches_per_second": 20.10758224451457,
+    "gpu_batches_per_second": 1.0414000868640614,
+    "producer_headroom": 19.308220249014926,
     "sustainable": true,
-    "buffer_seconds": 60.778249279716434,
+    "buffer_seconds": 60.49548179865254,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370687.3594653847,
-    "upload_bytes_per_second": 5762316.0177570805,
+    "generation_bytes_per_second": 402098.70782504015,
+    "upload_bytes_per_second": 5766497.401627052,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.54087818118244,
+    "estimated_full_shards_per_hour": 86.28102232039836,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 8470279,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9793791169995529,
+      "bytes": 5849196,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -51,3 +51,4 @@
 {"step": 562200, "loss": 0.5662493109703064, "lr": 0.0003, "grad_norm": 0.14734996078220597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82288728817, "tokens_per_second": 152798.27289510635, "data_wait_s": 9.854999916569795e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 562300, "loss": 0.6506227850914001, "lr": 0.0003, "grad_norm": 0.17583326428775464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82303413225, "tokens_per_second": 152641.61204161932, "data_wait_s": 1.2600999980350025e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 562400, "loss": 0.5580971240997314, "lr": 0.0003, "grad_norm": 0.1548995965877053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82318053862, "tokens_per_second": 152451.4765652134, "data_wait_s": 1.5704000361438375e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 562500, "loss": 0.6309000253677368, "lr": 0.0003, "grad_norm": 0.15265571888679594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82332737079, "tokens_per_second": 153379.97200531844, "data_wait_s": 1.0342000678065233e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:245a95748e694856d2bb3526bc6964c91b9de542bb09dbd7f309d5484772ea89
-size 113202
+oid sha256:c7dfbd409e48c7b792b1ee9e44e0b57b15c4b949b066dd22219db3c28378c230
+size 115320