Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75494.19472094062,
-      "required_pages_per_second_upper_estimate": 0.006543645460123913
+      "required_token_positions_per_second": 75499.19055677083,
+      "required_pages_per_second_upper_estimate": 0.006544078486511845
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9436.774340117578,
-      "required_pages_per_second_upper_estimate": 0.0037644703766226174
+      "required_token_positions_per_second": 9437.398819596354,
+      "required_pages_per_second_upper_estimate": 0.0037647194908235018
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18873.548680235155,
-      "required_pages_per_second_upper_estimate": 0.00436606332406041
+      "required_token_positions_per_second": 18874.797639192708,
+      "required_pages_per_second_upper_estimate": 0.004366352248734313
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28310.32302035273,
-      "required_pages_per_second_upper_estimate": 0.001723950139153812
+      "required_token_positions_per_second": 28312.19645878906,
+      "required_pages_per_second_upper_estimate": 0.0017240642217254127
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28310.32302035273,
-      "required_pages_per_second_upper_estimate": 0.002179407034096648
+      "required_token_positions_per_second": 28312.19645878906,
+      "required_pages_per_second_upper_estimate": 0.0021795512565734865
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28310.32302035273,
-      "required_pages_per_second_upper_estimate": 0.35151946335662776
+      "required_token_positions_per_second": 28312.19645878906,
+      "required_pages_per_second_upper_estimate": 0.3515427251920119
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09652076289847855,
-    "producer_p95_seconds": 0.2859838789991045,
-    "gpu_update_seconds": 0.9766048935620912,
-    "producer_batches_per_second": 10.360465147294882,
-    "gpu_batches_per_second": 1.0239555490578969,
-    "producer_headroom": 10.118080962427674,
+    "producer_batch_seconds": 0.08368022696876665,
+    "producer_p95_seconds": 0.2317200790021161,
+    "gpu_update_seconds": 0.9765402709127192,
+    "producer_batches_per_second": 11.950254393708164,
+    "gpu_batches_per_second": 1.0240233094180071,
+    "producer_headroom": 11.669904663107683,
     "sustainable": true,
-    "buffer_seconds": 60.54950340084966,
+    "buffer_seconds": 60.545496796588594,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450238.6981928125,
-    "upload_bytes_per_second": 5998205.232526224,
+    "generation_bytes_per_second": 450056.12764852785,
+    "upload_bytes_per_second": 5971105.842689773,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.61074361170083,
+    "estimated_full_shards_per_hour": 96.5715682229221,
     "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": 1.0888818870007526,
-      "bytes": 7323944,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.025877304999085,
       "bytes": 6206617,
@@ -562,6 +556,12 @@
       "bytes": 2553358,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2285513930000889,
+      "bytes": 6249359,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 97900, "loss": 0.6053578369319439, "lr": 0.0003, "grad_norm": 0.13006644836666734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14326096975, "tokens_per_second": 149703.34037703852, "data_wait_s": 1.416500163031742e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98000, "loss": 0.6172395683825016, "lr": 0.0003, "grad_norm": 0.13663269840799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14340795381, "tokens_per_second": 149430.98382918903, "data_wait_s": 1.4694000128656626e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98100, "loss": 0.6145087722688913, "lr": 0.0003, "grad_norm": 0.22264684725142575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14355414910, "tokens_per_second": 148326.88811174757, "data_wait_s": 1.3272001524455845e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 98200, "loss": 0.6513643469661474, "lr": 0.0003, "grad_norm": 0.12758982647775435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14369909417, "tokens_per_second": 146997.03627211528, "data_wait_s": 1.3330001820577309e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4290e2d34d0273e50cf5f41d58223ed7bcdb31e8ab06bafe311316e90aa95e71
-size 279189
+oid sha256:046742c0ff960ad92acb69aecfb3351af9d6b1ad9067445e3866bc6c0c6d8c35
+size 281307