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": 75420.47049911648,
-      "required_pages_per_second_upper_estimate": 0.005217763148730676
+      "required_token_positions_per_second": 76171.3299047943,
+      "required_pages_per_second_upper_estimate": 0.005269709344649322
     },
     {
       "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": 9427.55881238956,
-      "required_pages_per_second_upper_estimate": 0.003947122004932693
+      "required_token_positions_per_second": 9521.416238099288,
+      "required_pages_per_second_upper_estimate": 0.003986418149111432
     },
     {
       "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": 18855.11762477912,
-      "required_pages_per_second_upper_estimate": 0.003927023758351589
+      "required_token_positions_per_second": 19042.832476198575,
+      "required_pages_per_second_upper_estimate": 0.003966119811528732
     },
     {
       "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": 28282.67643716868,
-      "required_pages_per_second_upper_estimate": 0.0016955032672618438
+      "required_token_positions_per_second": 28564.24871429786,
+      "required_pages_per_second_upper_estimate": 0.0017123830953397652
     },
     {
       "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": 28282.67643716868,
-      "required_pages_per_second_upper_estimate": 0.002198853690643463
+      "required_token_positions_per_second": 28564.24871429786,
+      "required_pages_per_second_upper_estimate": 0.002220744696684699
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28282.67643716868,
-      "required_pages_per_second_upper_estimate": 0.3627798057640189
+      "required_token_positions_per_second": 28564.24871429786,
+      "required_pages_per_second_upper_estimate": 0.3663915126062757
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07319158585931973,
-    "producer_p95_seconds": 0.24357151400181465,
-    "gpu_update_seconds": 0.9775595340639475,
-    "producer_batches_per_second": 13.662772684309404,
-    "gpu_batches_per_second": 1.0229556003026865,
-    "producer_headroom": 13.356173699295129,
+    "producer_batch_seconds": 0.11683048781264915,
+    "producer_p95_seconds": 0.24077063000004273,
+    "gpu_update_seconds": 0.967923234268744,
+    "producer_batches_per_second": 8.559409608933693,
+    "gpu_batches_per_second": 1.0331397827798705,
+    "producer_headroom": 8.284851432110067,
     "sustainable": true,
-    "buffer_seconds": 60.608691111964745,
+    "buffer_seconds": 60.01124052466213,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395255.1513323576,
-    "upload_bytes_per_second": 5392558.667859726,
+    "generation_bytes_per_second": 397248.9891554783,
+    "upload_bytes_per_second": 5384716.570663779,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.81255440691038,
+    "estimated_full_shards_per_hour": 85.24038558958303,
     "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": 1.1176546409988077,
-      "bytes": 5812216,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.8849807390015485,
       "bytes": 1252571,
@@ -562,6 +556,12 @@
       "bytes": 6686271,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3011887590000697,
+      "bytes": 6177443,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 579400, "loss": 0.5096331238746643, "lr": 0.0003, "grad_norm": 0.1611796267319173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84806918627, "tokens_per_second": 150879.75185140574, "data_wait_s": 1.630000042496249e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579500, "loss": 0.6025071740150452, "lr": 0.0003, "grad_norm": 0.16887081732719386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84821567494, "tokens_per_second": 145632.65346693678, "data_wait_s": 1.22540004667826e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579600, "loss": 0.5668318271636963, "lr": 0.0003, "grad_norm": 0.16806697426411693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84836347216, "tokens_per_second": 153372.8027465118, "data_wait_s": 8.847000572131947e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 579700, "loss": 0.5969656109809875, "lr": 0.0003, "grad_norm": 0.13797264914020052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84850919354, "tokens_per_second": 152031.74969802043, "data_wait_s": 9.548999514663592e-06, "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:b59c64e55860a2d21d8dfb23aed02b8e648a713f0898a412df71b9cb6fd4c016
-size 479847
+oid sha256:92b81d7e2cf1234541796689a49692705e45df12fe202e2b24b9bbce68d64566
+size 481965