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": 76207.84535760675,
-      "required_pages_per_second_upper_estimate": 0.005272235568402411
+      "required_token_positions_per_second": 76399.03700540696,
+      "required_pages_per_second_upper_estimate": 0.005285462650222972
     },
     {
       "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": 9525.980669700844,
-      "required_pages_per_second_upper_estimate": 0.003988329181306833
+      "required_token_positions_per_second": 9549.87962567587,
+      "required_pages_per_second_upper_estimate": 0.003998335175106625
     },
     {
       "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": 19051.96133940169,
-      "required_pages_per_second_upper_estimate": 0.0039680211129896285
+      "required_token_positions_per_second": 19099.75925135174,
+      "required_pages_per_second_upper_estimate": 0.00397797615753312
     },
     {
       "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": 28577.942009102535,
-      "required_pages_per_second_upper_estimate": 0.0017132039874548548
+      "required_token_positions_per_second": 28649.638877027613,
+      "required_pages_per_second_upper_estimate": 0.0017175021052121848
     },
     {
       "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": 28577.942009102535,
-      "required_pages_per_second_upper_estimate": 0.002221809290125324
+      "required_token_positions_per_second": 28649.638877027613,
+      "required_pages_per_second_upper_estimate": 0.002227383406245305
     },
     {
       "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": 28577.942009102535,
-      "required_pages_per_second_upper_estimate": 0.3665671554893156
+      "required_token_positions_per_second": 28649.638877027613,
+      "required_pages_per_second_upper_estimate": 0.36748680592895955
     }
   ],
   "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": 0.04693305085939059,
-    "producer_p95_seconds": 0.15705347199991593,
-    "gpu_update_seconds": 0.9674594479614266,
-    "producer_batches_per_second": 21.306946420252057,
-    "gpu_batches_per_second": 1.0336350553060814,
-    "producer_headroom": 20.61360662148075,
+    "producer_batch_seconds": 0.06976582499218864,
+    "producer_p95_seconds": 0.15564646299981177,
+    "gpu_update_seconds": 0.9650383419725836,
+    "producer_batches_per_second": 14.333665517636542,
+    "gpu_batches_per_second": 1.0362282579943436,
+    "producer_headroom": 13.832536805529564,
     "sustainable": true,
-    "buffer_seconds": 60.94994522156988,
+    "buffer_seconds": 60.797415544272766,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288681.81567301677,
+    "generation_bytes_per_second": 321809.446246533,
     "upload_bytes_per_second": 4960346.038410955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.94439747469785,
+    "estimated_full_shards_per_hour": 69.05281582400315,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -25,3 +25,4 @@
 {"step": 606900, "loss": 0.5948517918586731, "lr": 0.0001, "grad_norm": 0.1392813678864415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88832465197, "tokens_per_second": 153589.1441208933, "data_wait_s": 1.406699993822258e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 607000, "loss": 0.5789181590080261, "lr": 0.0001, "grad_norm": 0.13183917411198898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88847190428, "tokens_per_second": 152226.69748541186, "data_wait_s": 9.513999884802615e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 607100, "loss": 0.5816693305969238, "lr": 0.0001, "grad_norm": 0.1479278366716872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88861688463, "tokens_per_second": 150020.00815992078, "data_wait_s": 1.2137999874539673e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 607200, "loss": 0.5858821272850037, "lr": 0.0001, "grad_norm": 0.17613203362462435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88876154434, "tokens_per_second": 150519.37055730794, "data_wait_s": 1.4258999726735055e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8617d5bf4afd0063851d1d255e835e78a3cc1639edda04ff996b99833a953af
-size 58134
+oid sha256:3c95a3e7e604d2f8c961303a701f8c45ab15129e0fb7840e3ebcd1f2f5835916
+size 60252