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": 76892.80021374699,
-      "required_pages_per_second_upper_estimate": 0.005319622334664423
+      "required_token_positions_per_second": 75955.07204125375,
+      "required_pages_per_second_upper_estimate": 0.0052547481238622105
     },
     {
       "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": 9611.600026718374,
-      "required_pages_per_second_upper_estimate": 0.0040241762181545855
+      "required_token_positions_per_second": 9494.384005156719,
+      "required_pages_per_second_upper_estimate": 0.003975100317675594
     },
     {
       "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": 19223.200053436747,
-      "required_pages_per_second_upper_estimate": 0.004003685621254562
+      "required_token_positions_per_second": 18988.768010313437,
+      "required_pages_per_second_upper_estimate": 0.003954859609060702
     },
     {
       "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": 28834.80008015512,
-      "required_pages_per_second_upper_estimate": 0.001728602236614892
+      "required_token_positions_per_second": 28483.152015470154,
+      "required_pages_per_second_upper_estimate": 0.0017075214720725334
     },
     {
       "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": 28834.80008015512,
-      "required_pages_per_second_upper_estimate": 0.0022417788753504087
+      "required_token_positions_per_second": 28483.152015470154,
+      "required_pages_per_second_upper_estimate": 0.002214439785057525
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28834.80008015512,
-      "required_pages_per_second_upper_estimate": 0.36986185503206886
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28483.152015470154,
+      "required_pages_per_second_upper_estimate": 0.36535129122856497
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06212936620308085,
-    "producer_p95_seconds": 0.21145619900016754,
-    "gpu_update_seconds": 0.9588413973096381,
-    "producer_batches_per_second": 16.095448273708808,
-    "gpu_batches_per_second": 1.0429253501213513,
-    "producer_headroom": 15.432982113087956,
+    "producer_batch_seconds": 0.6579235090234903,
+    "producer_p95_seconds": 0.23980363800001214,
+    "gpu_update_seconds": 0.970679087236674,
+    "producer_batches_per_second": 1.5199335276592105,
+    "gpu_batches_per_second": 1.030206597781762,
+    "producer_headroom": 1.4753676892886605,
     "sustainable": true,
-    "buffer_seconds": 60.4070080305072,
+    "buffer_seconds": 60.18210340867378,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426191.49670901796,
-    "upload_bytes_per_second": 4707018.768722284,
+    "generation_bytes_per_second": 361278.9601964715,
+    "upload_bytes_per_second": 4654056.376878685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.45077396347908,
+    "estimated_full_shards_per_hour": 77.52205471439942,
     "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": 1.0616652130001967,
-      "bytes": 6074717,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3232113710000704,
       "bytes": 5040655,
@@ -562,6 +556,12 @@
       "bytes": 7705972,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9795733470000414,
+      "bytes": 923741,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 855500, "loss": 0.467849999666214, "lr": 3e-05, "grad_norm": 0.31774467713202603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125221109740, "tokens_per_second": 152338.72281623568, "data_wait_s": 1.394600076309871e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 855600, "loss": 0.5046199560165405, "lr": 3e-05, "grad_norm": 0.2905278460675706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125235847834, "tokens_per_second": 153236.85802427933, "data_wait_s": 1.7033000403898768e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 855700, "loss": 0.508650541305542, "lr": 3e-05, "grad_norm": 0.3107761555130699, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125250457357, "tokens_per_second": 141593.71435859895, "data_wait_s": 7.3999999585794285e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 855800, "loss": 0.5231516361236572, "lr": 3e-05, "grad_norm": 0.2864965672270696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125265107838, "tokens_per_second": 152522.0216723379, "data_wait_s": 1.245400017069187e-05, "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 sha256:86f314b0b00ade63c39c5a40ef2c9509867b74f64e77fcc5f0ff33638afc8ed2
-size 282090
+oid sha256:20379632cfba0d6520c9807370208e358f2fd9d7e0dd4a8fd8f1e61c86f9ec7d
+size 284208