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": 76725.12463730051,
-      "required_pages_per_second_upper_estimate": 0.005308022149224917
+      "required_token_positions_per_second": 75882.55821766844,
+      "required_pages_per_second_upper_estimate": 0.005249731449291327
     },
     {
       "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": 9590.640579662564,
-      "required_pages_per_second_upper_estimate": 0.004015400935355343
+      "required_token_positions_per_second": 9485.319777208555,
+      "required_pages_per_second_upper_estimate": 0.003971305314716301
     },
     {
       "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": 19181.281159325128,
-      "required_pages_per_second_upper_estimate": 0.00399495502108669
+      "required_token_positions_per_second": 18970.63955441711,
+      "required_pages_per_second_upper_estimate": 0.003951083929777028
     },
     {
       "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": 28771.92173898769,
-      "required_pages_per_second_upper_estimate": 0.0017248327760715724
+      "required_token_positions_per_second": 28455.95933162567,
+      "required_pages_per_second_upper_estimate": 0.0017058913121968807
     },
     {
       "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": 28771.92173898769,
-      "required_pages_per_second_upper_estimate": 0.002236890360897235
+      "required_token_positions_per_second": 28455.95933162567,
+      "required_pages_per_second_upper_estimate": 0.002212325673496592
     },
     {
       "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": 28771.92173898769,
-      "required_pages_per_second_upper_estimate": 0.3690553191850757
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28455.95933162567,
+      "required_pages_per_second_upper_estimate": 0.36500249267743706
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 704566,
-    "producer_batch_seconds": 0.09442553102341122,
-    "producer_p95_seconds": 0.15989514000011695,
-    "gpu_update_seconds": 0.9609368554112008,
-    "producer_batches_per_second": 10.590356116208303,
-    "gpu_batches_per_second": 1.0406511045640803,
-    "producer_headroom": 10.176663503993984,
+    "producer_batch_seconds": 0.18798613368748818,
+    "producer_p95_seconds": 0.15358469900002092,
+    "gpu_update_seconds": 0.9716066739409587,
+    "producer_batches_per_second": 5.319541289478401,
+    "gpu_batches_per_second": 1.029223066103359,
+    "producer_headroom": 5.168501819161708,
     "sustainable": true,
-    "buffer_seconds": 60.53902189090565,
+    "buffer_seconds": 60.23961378433944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 559218.7652134837,
-    "upload_bytes_per_second": 4198592.797013168,
+    "generation_bytes_per_second": 519474.457935601,
+    "upload_bytes_per_second": 4247862.740627061,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.99532906821617,
+    "estimated_full_shards_per_hour": 111.46712592650435,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 6096975,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.174733095999727,
+      "bytes": 7198373,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 443800, "loss": 0.5933117270469666, "lr": 0.0003, "grad_norm": 0.17910241468328222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64959227114, "tokens_per_second": 153051.33207088735, "data_wait_s": 1.30259995785309e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 443900, "loss": 0.5994600653648376, "lr": 0.0003, "grad_norm": 0.1544873868135243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64973887332, "tokens_per_second": 152285.32138178108, "data_wait_s": 1.5888000234554056e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444000, "loss": 0.6055135726928711, "lr": 0.0003, "grad_norm": 0.17247214597601487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64988420632, "tokens_per_second": 150676.11953123167, "data_wait_s": 1.2222999430377968e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 444100, "loss": 0.5636587142944336, "lr": 0.0003, "grad_norm": 0.1544770824898031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65002947538, "tokens_per_second": 150526.41377154645, "data_wait_s": 1.225300002261065e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f3696e26de04e1bdcb75c1a2de9c5792c14f7dbbc9b2a938856846ff6d73206
-size 80649
+oid sha256:fa44a1ec0d5af78b90ac6e5658c0835c1ac33380d1a2662aeb81a573a09b0e82
+size 82767