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": 77320.73782381108,
-      "required_pages_per_second_upper_estimate": 0.005349228051480715
+      "required_token_positions_per_second": 77299.54458622036,
+      "required_pages_per_second_upper_estimate": 0.005347761854129103
     },
     {
       "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": 9665.092227976385,
-      "required_pages_per_second_upper_estimate": 0.004046572285777129
+      "required_token_positions_per_second": 9662.443073277545,
+      "required_pages_per_second_upper_estimate": 0.0040454631400253654
     },
     {
       "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": 19330.18445595277,
-      "required_pages_per_second_upper_estimate": 0.004025967650930227
+      "required_token_positions_per_second": 19324.88614655509,
+      "required_pages_per_second_upper_estimate": 0.004024864152808506
     },
     {
       "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": 28995.276683929154,
-      "required_pages_per_second_upper_estimate": 0.0017382225639775593
+      "required_token_positions_per_second": 28987.329219832638,
+      "required_pages_per_second_upper_estimate": 0.0017377461256400481
     },
     {
       "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": 28995.276683929154,
-      "required_pages_per_second_upper_estimate": 0.002254255225431858
+      "required_token_positions_per_second": 28987.329219832638,
+      "required_pages_per_second_upper_estimate": 0.0022536373450555537
     },
     {
       "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": 28995.276683929154,
-      "required_pages_per_second_upper_estimate": 0.3719202765989296
+      "required_token_positions_per_second": 28987.329219832638,
+      "required_pages_per_second_upper_estimate": 0.37181833506282164
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07411278338277327,
-    "producer_p95_seconds": 0.20268456099984178,
-    "gpu_update_seconds": 0.9535346153576837,
-    "producer_batches_per_second": 13.492948913215953,
-    "gpu_batches_per_second": 1.0487296254314653,
-    "producer_headroom": 12.86599385200425,
+    "producer_batch_seconds": 0.0753484537891893,
+    "producer_p95_seconds": 0.20457571400038432,
+    "gpu_update_seconds": 0.953796046207793,
+    "producer_batches_per_second": 13.271672472507671,
+    "gpu_batches_per_second": 1.0484421737497338,
+    "producer_headroom": 12.658468730842621,
     "sustainable": true,
-    "buffer_seconds": 60.072680767534074,
+    "buffer_seconds": 60.08915091109096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339208.2126583265,
+    "generation_bytes_per_second": 387390.8187987446,
     "upload_bytes_per_second": 5429589.450332749,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.78618607342096,
+    "estimated_full_shards_per_hour": 83.1250517174888,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -119,3 +119,4 @@
 {"step": 569000, "loss": 0.5846968293190002, "lr": 0.0003, "grad_norm": 0.16389304236523466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83284694450, "tokens_per_second": 150266.70682100102, "data_wait_s": 1.3680000847671181e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 569100, "loss": 0.5906668305397034, "lr": 0.0003, "grad_norm": 0.15811197033862542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83299438059, "tokens_per_second": 153795.93565720643, "data_wait_s": 8.357001206604764e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 569200, "loss": 0.6039405465126038, "lr": 0.0003, "grad_norm": 0.1723336306602568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83314124675, "tokens_per_second": 153280.5835935192, "data_wait_s": 1.2098998922738247e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 569300, "loss": 0.6279890537261963, "lr": 0.0003, "grad_norm": 0.17659084689620977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83328875507, "tokens_per_second": 153471.89007755465, "data_wait_s": 1.1825000910903327e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3494182d902f439865045201d59fa14b62d1a679209dc3c72686110bd4ac5ea9
-size 258009
+oid sha256:46750354d428c0919f2b138e83a8f348e89a7b3b116acc6812c6493bf4d51911
+size 260127