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": 76272.35576174624,
-      "required_pages_per_second_upper_estimate": 0.006611094486909816
+      "required_token_positions_per_second": 75244.40110439017,
+      "required_pages_per_second_upper_estimate": 0.006521993982537453
     },
     {
       "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": 9534.04447021828,
-      "required_pages_per_second_upper_estimate": 0.0038032728858378327
+      "required_token_positions_per_second": 9405.550138048771,
+      "required_pages_per_second_upper_estimate": 0.0037520145755739476
     },
     {
       "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": 19068.08894043656,
-      "required_pages_per_second_upper_estimate": 0.004411066789466362
+      "required_token_positions_per_second": 18811.100276097543,
+      "required_pages_per_second_upper_estimate": 0.004351616984817549
     },
     {
       "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": 28602.13341065484,
-      "required_pages_per_second_upper_estimate": 0.0017417198609124152
+      "required_token_positions_per_second": 28216.650414146316,
+      "required_pages_per_second_upper_estimate": 0.0017182459688980238
     },
     {
       "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": 28602.13341065484,
-      "required_pages_per_second_upper_estimate": 0.002201871405724965
+      "required_token_positions_per_second": 28216.650414146316,
+      "required_pages_per_second_upper_estimate": 0.002172195857568508
     },
     {
       "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": 28602.13341065484,
-      "required_pages_per_second_upper_estimate": 0.35514277177762815
+      "required_token_positions_per_second": 28216.650414146316,
+      "required_pages_per_second_upper_estimate": 0.35035636308959006
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10577365055451082,
-    "producer_p95_seconds": 0.40235264799412107,
-    "gpu_update_seconds": 0.9666411803288979,
-    "producer_batches_per_second": 9.454150393387874,
-    "gpu_batches_per_second": 1.0345100336608377,
-    "producer_headroom": 9.13877109527137,
+    "producer_batch_seconds": 0.16650059692193508,
+    "producer_p95_seconds": 0.4007023480007774,
+    "gpu_update_seconds": 0.9798469908440577,
+    "producer_batches_per_second": 6.005984473850606,
+    "gpu_batches_per_second": 1.0205675062986947,
+    "producer_headroom": 5.884945813758646,
     "sustainable": true,
-    "buffer_seconds": 60.898394360720566,
+    "buffer_seconds": 60.75051343233157,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 245737.31614883497,
-    "upload_bytes_per_second": 5889720.439706964,
+    "generation_bytes_per_second": 243433.6254374861,
+    "upload_bytes_per_second": 5948986.567524921,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.72950757359301,
+    "estimated_full_shards_per_hour": 52.235189174112676,
     "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": 1.0104034229989338,
-      "bytes": 6117004,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.275026989998878,
       "bytes": 16016482,
@@ -562,6 +556,12 @@
       "bytes": 973375,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.5195651320027537,
+      "bytes": 13317753,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -380,3 +380,4 @@
 {"step": 153000, "loss": 0.6109720841050148, "lr": 0.0003, "grad_norm": 0.1675781607171175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22391140179, "tokens_per_second": 147429.25054257218, "data_wait_s": 9.397001122124493e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 153100, "loss": 0.5912684127688408, "lr": 0.0003, "grad_norm": 0.14043988631070992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22405728801, "tokens_per_second": 149638.74641906188, "data_wait_s": 9.56199801294133e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 153200, "loss": 0.5679905088618398, "lr": 0.0003, "grad_norm": 0.15123643265815698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22420385300, "tokens_per_second": 119521.99032782183, "data_wait_s": 1.469699782319367e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 153300, "loss": 0.6331153865903616, "lr": 0.0003, "grad_norm": 0.18648370077675916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22435097611, "tokens_per_second": 150503.97369121175, "data_wait_s": 1.3190998288337141e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bde38726f9fdd87c421852b23bae38e1176488446c073c7e07714fba104a2a7b
-size 814722
+oid sha256:ab65caca80bdae06c2b77c866eee981b7403951535a180b09c601ae2c2d86ae0
+size 816840