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": 76364.49827648082,
-      "required_pages_per_second_upper_estimate": 0.005283073180814969
+      "required_token_positions_per_second": 77071.50823694911,
+      "required_pages_per_second_upper_estimate": 0.005331985770369289
     },
     {
       "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": 9545.562284560103,
-      "required_pages_per_second_upper_estimate": 0.003996527594537788
+      "required_token_positions_per_second": 9633.93852961864,
+      "required_pages_per_second_upper_estimate": 0.0040335288828379405
     },
     {
       "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": 19091.124569120206,
-      "required_pages_per_second_upper_estimate": 0.003976177780936174
+      "required_token_positions_per_second": 19267.87705923728,
+      "required_pages_per_second_upper_estimate": 0.004012990663350939
     },
     {
       "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": 28636.686853680312,
-      "required_pages_per_second_upper_estimate": 0.0017167256511891086
+      "required_token_positions_per_second": 28901.81558885592,
+      "required_pages_per_second_upper_estimate": 0.0017326197140347481
     },
     {
       "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": 28636.686853680312,
-      "required_pages_per_second_upper_estimate": 0.0022263764433999825
+      "required_token_positions_per_second": 28901.81558885592,
+      "required_pages_per_second_upper_estimate": 0.0022469890363818255
     },
     {
       "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": 28636.686853680312,
-      "required_pages_per_second_upper_estimate": 0.3673206712802595
+      "required_token_positions_per_second": 28901.81558885592,
+      "required_pages_per_second_upper_estimate": 0.3707214580220356
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10391373147653837,
-    "producer_p95_seconds": 0.18693384500147658,
-    "gpu_update_seconds": 0.965474817015948,
-    "producer_batches_per_second": 9.623367246953112,
-    "gpu_batches_per_second": 1.0357597965017473,
-    "producer_headroom": 9.291118731829322,
+    "producer_batch_seconds": 0.05645111726566654,
+    "producer_p95_seconds": 0.18069455300064874,
+    "gpu_update_seconds": 0.9566181029353958,
+    "producer_batches_per_second": 17.71444124469434,
+    "gpu_batches_per_second": 1.0453492328145226,
+    "producer_headroom": 16.94595517806003,
     "sustainable": true,
-    "buffer_seconds": 60.824913472004724,
+    "buffer_seconds": 60.26694048492994,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324462.2512382495,
-    "upload_bytes_per_second": 5607043.908305009,
+    "generation_bytes_per_second": 324588.2194164283,
+    "upload_bytes_per_second": 5560784.731990927,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.62204602108586,
+    "estimated_full_shards_per_hour": 69.64907585973393,
     "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": 1.233521140999983,
-      "bytes": 10999051,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 1.1022731230000318,
       "bytes": 7378325,
@@ -562,6 +556,12 @@
       "bytes": 5904691,
       "pages": 43,
       "shards": 2
+    },
+    {
+      "seconds": 1.2678156700003456,
+      "bytes": 7637773,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -92,3 +92,4 @@
 {"step": 613600, "loss": 0.5718985199928284, "lr": 0.0001, "grad_norm": 0.15181234662388043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89812861909, "tokens_per_second": 152798.6998932967, "data_wait_s": 1.014500048768241e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 613700, "loss": 0.5262578725814819, "lr": 0.0001, "grad_norm": 0.15743298789531332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89827567449, "tokens_per_second": 153356.35091314453, "data_wait_s": 1.507399974798318e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 613800, "loss": 0.5992052555084229, "lr": 0.0001, "grad_norm": 0.14791244464484515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89842313948, "tokens_per_second": 153420.90148336272, "data_wait_s": 1.2147000234108418e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 613900, "loss": 0.5253154635429382, "lr": 0.0001, "grad_norm": 0.15046249175064386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89857060438, "tokens_per_second": 153234.40435261215, "data_wait_s": 1.4196999472915195e-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 sha256:7488ea9cbdaf7ef498be6b9b2a8644212b47c2c4165bbd34a6bdc3f59b421ed8
-size 200823
+oid sha256:b62a85a41288f46449b6aca22737c1bf0b81bbec0601a6e95a3066f195d97be7
+size 202941