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": 76987.70196799481,
-      "required_pages_per_second_upper_estimate": 0.005326187858225696
+      "required_token_positions_per_second": 76438.11963283618,
+      "required_pages_per_second_upper_estimate": 0.005288166477072715
     },
     {
       "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": 9623.462745999352,
-      "required_pages_per_second_upper_estimate": 0.004029142890995783
+      "required_token_positions_per_second": 9554.764954104523,
+      "required_pages_per_second_upper_estimate": 0.004000380560102444
     },
     {
       "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": 19246.925491998703,
-      "required_pages_per_second_upper_estimate": 0.004008627004425128
+      "required_token_positions_per_second": 19109.529908209046,
+      "required_pages_per_second_upper_estimate": 0.003980011127686947
     },
     {
       "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": 28870.388237998057,
-      "required_pages_per_second_upper_estimate": 0.0017307356923376054
+      "required_token_positions_per_second": 28664.294862313567,
+      "required_pages_per_second_upper_estimate": 0.001718380709151684
     },
     {
       "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": 28870.388237998057,
-      "required_pages_per_second_upper_estimate": 0.002244545698086934
+      "required_token_positions_per_second": 28664.294862313567,
+      "required_pages_per_second_upper_estimate": 0.0022285228446364217
     },
     {
       "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": 28870.388237998057,
-      "required_pages_per_second_upper_estimate": 0.3703183417092913
+      "required_token_positions_per_second": 28664.294862313567,
+      "required_pages_per_second_upper_estimate": 0.36767479717183676
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1136449706640974,
-    "producer_p95_seconds": 0.3292903820038191,
-    "gpu_update_seconds": 0.9576594457988896,
-    "producer_batches_per_second": 8.799333522252551,
-    "gpu_batches_per_second": 1.0442125375433324,
-    "producer_headroom": 8.426764864319969,
+    "producer_batch_seconds": 0.16815493842989326,
+    "producer_p95_seconds": 0.3296454539959086,
+    "gpu_update_seconds": 0.9645449201804805,
+    "producer_batches_per_second": 5.946896411947589,
+    "gpu_batches_per_second": 1.0367583500547441,
+    "producer_headroom": 5.736048724983573,
     "sustainable": true,
-    "buffer_seconds": 60.33254508533005,
+    "buffer_seconds": 60.766329971370276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 274657.8927608813,
-    "upload_bytes_per_second": 4912467.421172711,
+    "generation_bytes_per_second": 274978.6344108315,
+    "upload_bytes_per_second": 4965245.368220504,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.93519007797078,
+    "estimated_full_shards_per_hour": 59.00401376956662,
     "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": 1.5255058740003733,
-      "bytes": 7690292,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.0536525450006593,
       "bytes": 865655,
@@ -562,6 +556,12 @@
       "bytes": 5519624,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 1.6776565620020847,
+      "bytes": 13016413,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -379,3 +379,4 @@
 {"step": 595000, "loss": 0.5304991602897644, "lr": 0.0003, "grad_norm": 0.16238917874349468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87090245215, "tokens_per_second": 151130.23713334586, "data_wait_s": 1.0615003702696413e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 595100, "loss": 0.6110762357711792, "lr": 0.0003, "grad_norm": 0.17963012212375318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87104740682, "tokens_per_second": 139207.420233886, "data_wait_s": 1.4395998732652515e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 595200, "loss": 0.5996808409690857, "lr": 0.0003, "grad_norm": 0.1724937067580471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87119130921, "tokens_per_second": 150117.92940040177, "data_wait_s": 1.0970004950650036e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 595300, "loss": 0.5223484635353088, "lr": 0.0003, "grad_norm": 0.16592439938990478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87133793712, "tokens_per_second": 152539.51282323984, "data_wait_s": 1.0310002835467458e-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 sha256:0a8dcb2090b6dee57be5d72586a6fcd5a29aa268e2299d28ca4d380a1776df2e
-size 813387
+oid sha256:567c8fa2461bcb591abcc8e9230b163ccef982037a410667e86b0a6d302a62e5
+size 815505