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": 76266.35401620071,
-      "required_pages_per_second_upper_estimate": 0.005276283333162733
+      "required_token_positions_per_second": 76907.92586092695,
+      "required_pages_per_second_upper_estimate": 0.005320668762032669
     },
     {
       "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": 9533.294252025089,
-      "required_pages_per_second_upper_estimate": 0.003991391225501029
+      "required_token_positions_per_second": 9613.490732615868,
+      "required_pages_per_second_upper_estimate": 0.004024967817231438
     },
     {
       "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": 19066.588504050178,
-      "required_pages_per_second_upper_estimate": 0.003971067565641637
+      "required_token_positions_per_second": 19226.981465231736,
+      "required_pages_per_second_upper_estimate": 0.004004473189608924
     },
     {
       "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": 28599.882756075265,
-      "required_pages_per_second_upper_estimate": 0.0017145193017342893
+      "required_token_positions_per_second": 28840.4721978476,
+      "required_pages_per_second_upper_estimate": 0.001728942271409734
     },
     {
       "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": 28599.882756075265,
-      "required_pages_per_second_upper_estimate": 0.002223515086695307
+      "required_token_positions_per_second": 28840.4721978476,
+      "required_pages_per_second_upper_estimate": 0.00224221985755199
     },
     {
       "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": 28599.882756075265,
-      "required_pages_per_second_upper_estimate": 0.3668485878333432
+      "required_token_positions_per_second": 28840.4721978476,
+      "required_pages_per_second_upper_estimate": 0.3699346108675825
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06440008534367792,
-    "producer_p95_seconds": 0.20835253299992473,
-    "gpu_update_seconds": 0.9667172497106457,
-    "producer_batches_per_second": 15.527929732754133,
-    "gpu_batches_per_second": 1.0344286297770278,
-    "producer_headroom": 15.011117524948236,
+    "producer_batch_seconds": 0.08299708391405147,
+    "producer_p95_seconds": 0.23465608100013924,
+    "gpu_update_seconds": 0.9586528199099112,
+    "producer_batches_per_second": 12.048616081928381,
+    "gpu_batches_per_second": 1.0431305048411315,
+    "producer_headroom": 11.55043978295255,
     "sustainable": true,
-    "buffer_seconds": 60.903186731770674,
+    "buffer_seconds": 60.39512765432441,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452898.3461550163,
+    "generation_bytes_per_second": 454647.7291874037,
     "upload_bytes_per_second": 4789721.178896227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.18144215095393,
+    "estimated_full_shards_per_hour": 97.5568190261515,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -120,3 +120,4 @@
 {"step": 854500, "loss": 0.5277312397956848, "lr": 3e-05, "grad_norm": 0.30865370008704085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125074912455, "tokens_per_second": 153124.75779856162, "data_wait_s": 1.2423999578459188e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 854600, "loss": 0.5045964121818542, "lr": 3e-05, "grad_norm": 0.30025854199019913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125089655615, "tokens_per_second": 153482.20194672578, "data_wait_s": 1.1187999916728586e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 854700, "loss": 0.5163973569869995, "lr": 3e-05, "grad_norm": 0.2944249649509857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125104272160, "tokens_per_second": 152227.42691248658, "data_wait_s": 1.465599962102715e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 854800, "loss": 0.5010818243026733, "lr": 3e-05, "grad_norm": 0.312108886562663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125118782654, "tokens_per_second": 150846.80249759907, "data_wait_s": 6.986998414504342e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0def1cd4933aeaad0c85f37b1971487367714bacf5055db9658396c53f1fb21d
-size 260127
+oid sha256:107589e2c11a7b200f77b44e9d32149fa4a1da27cb7973fa9a8018ee35c6ff65
+size 262245