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": 76370.31081272103,
-      "required_pages_per_second_upper_estimate": 0.005283475305559009
+      "required_token_positions_per_second": 75712.28521139496,
+      "required_pages_per_second_upper_estimate": 0.005237951567629521
     },
     {
       "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": 9546.288851590129,
-      "required_pages_per_second_upper_estimate": 0.003996831792980815
+      "required_token_positions_per_second": 9464.03565142437,
+      "required_pages_per_second_upper_estimate": 0.003962394095713551
     },
     {
       "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": 19092.577703180257,
-      "required_pages_per_second_upper_estimate": 0.003976480430439161
+      "required_token_positions_per_second": 18928.07130284874,
+      "required_pages_per_second_upper_estimate": 0.003942218085575623
     },
     {
       "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": 28638.866554770386,
-      "required_pages_per_second_upper_estimate": 0.0017168563209412476
+      "required_token_positions_per_second": 28392.106954273113,
+      "required_pages_per_second_upper_estimate": 0.00170206345967153
     },
     {
       "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": 28638.866554770386,
-      "required_pages_per_second_upper_estimate": 0.002226545905572244
+      "required_token_positions_per_second": 28392.106954273113,
+      "required_pages_per_second_upper_estimate": 0.0022073614320143567
     },
     {
       "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": 28638.866554770386,
-      "required_pages_per_second_upper_estimate": 0.3673486301454623
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28392.106954273113,
+      "required_pages_per_second_upper_estimate": 0.3641834629400997
     }
   ],
   "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": 0.06226728067957765,
-    "producer_p95_seconds": 0.21873538699946948,
-    "gpu_update_seconds": 0.9654013348302244,
-    "producer_batches_per_second": 16.059798807433367,
-    "gpu_batches_per_second": 1.0358386340701093,
-    "producer_headroom": 15.50415120580102,
+    "producer_batch_seconds": 0.5506624080547056,
+    "producer_p95_seconds": 0.2382003700004134,
+    "gpu_update_seconds": 0.973791767005121,
+    "producer_batches_per_second": 1.8159946736379633,
+    "gpu_batches_per_second": 1.0269135906493458,
+    "producer_headroom": 1.7684006621138004,
     "sustainable": true,
-    "buffer_seconds": 60.820284094304135,
+    "buffer_seconds": 60.375089554317505,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420386.5206501806,
-    "upload_bytes_per_second": 3423798.4880936784,
+    "generation_bytes_per_second": 367646.6533307506,
+    "upload_bytes_per_second": 3393180.3976997966,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.20516123417914,
+    "estimated_full_shards_per_hour": 78.88841342870606,
     "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": 1.7619018259993027,
-      "bytes": 7674402,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.6144984039983683,
       "bytes": 4728422,
@@ -562,6 +556,12 @@
       "bytes": 302035,
       "pages": 9,
       "shards": 1
+    },
+    {
+      "seconds": 0.93341918599981,
+      "bytes": 1061744,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -175,3 +175,4 @@
 {"step": 860000, "loss": 0.5167421102523804, "lr": 3e-05, "grad_norm": 0.3017758321480747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125879763724, "tokens_per_second": 150859.77013761847, "data_wait_s": 1.5797999367350712e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 860100, "loss": 0.47436317801475525, "lr": 3e-05, "grad_norm": 0.30331582675117236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125894241221, "tokens_per_second": 144668.14853932388, "data_wait_s": 1.6124002286233008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 860200, "loss": 0.48211637139320374, "lr": 3e-05, "grad_norm": 0.3014901763401981, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125908829666, "tokens_per_second": 151120.25287623127, "data_wait_s": 1.681599678704515e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 860300, "loss": 0.5644816160202026, "lr": 3e-05, "grad_norm": 0.31252708845055804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125923497239, "tokens_per_second": 151458.2176241662, "data_wait_s": 1.382999835186638e-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:e3e3ab3a9f40d330c824da701b416e2107f7eab122d76822b5d31e856820aca5
-size 378183
+oid sha256:0b8455323ce6bca6c002416efb20aa30163b11325be31b0c5749e8352a770827
+size 380301