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": 75735.15381934233,
-      "required_pages_per_second_upper_estimate": 0.006564531184068121
+      "required_token_positions_per_second": 74313.08728854939,
+      "required_pages_per_second_upper_estimate": 0.00644127006137366
     },
     {
       "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": 9466.894227417792,
-      "required_pages_per_second_upper_estimate": 0.003776485649999119
+      "required_token_positions_per_second": 9289.135911068674,
+      "required_pages_per_second_upper_estimate": 0.0037055751998837853
     },
     {
       "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": 18933.788454835583,
-      "required_pages_per_second_upper_estimate": 0.004379998735730395
+      "required_token_positions_per_second": 18578.271822137347,
+      "required_pages_per_second_upper_estimate": 0.004297756219634696
     },
     {
       "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": 28400.682682253373,
-      "required_pages_per_second_upper_estimate": 0.001729452568483054
+      "required_token_positions_per_second": 27867.407733206022,
+      "required_pages_per_second_upper_estimate": 0.0016969789219634967
     },
     {
       "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": 28400.682682253373,
-      "required_pages_per_second_upper_estimate": 0.0021863631710013127
+      "required_token_positions_per_second": 27867.407733206022,
+      "required_pages_per_second_upper_estimate": 0.002145310189224108
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28400.682682253373,
-      "required_pages_per_second_upper_estimate": 0.3526414279430991
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27867.407733206022,
+      "required_pages_per_second_upper_estimate": 0.34601993783237545
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.26521474165636505,
-    "producer_p95_seconds": 0.3362971189999371,
-    "gpu_update_seconds": 0.9734977257175688,
-    "producer_batches_per_second": 3.770529472662895,
-    "gpu_batches_per_second": 1.027223765995854,
-    "producer_headroom": 3.670601866388392,
+    "producer_batch_seconds": 0.08049295572678261,
+    "producer_p95_seconds": 0.2895517329998256,
+    "gpu_update_seconds": 0.9921267261272895,
+    "producer_batches_per_second": 12.423447380840404,
+    "gpu_batches_per_second": 1.0079357542392224,
+    "producer_headroom": 12.325634177167839,
     "sustainable": true,
-    "buffer_seconds": 60.356858994489265,
+    "buffer_seconds": 60.51973029376466,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 274185.13017651666,
-    "upload_bytes_per_second": 4949851.157324105,
+    "generation_bytes_per_second": 295605.63327969244,
+    "upload_bytes_per_second": 5016680.774782323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.833746232715846,
+    "estimated_full_shards_per_hour": 63.43008755486564,
     "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": 1.8189289659967471,
-      "bytes": 7675879,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.221604631999071,
       "bytes": 830309,
@@ -562,6 +556,12 @@
       "bytes": 913950,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.2434416440009954,
+      "bytes": 10283141,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 260500, "loss": 0.6168870646506548, "lr": 0.0003, "grad_norm": 0.1660470359645184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38130363613, "tokens_per_second": 150256.56039720727, "data_wait_s": 9.6180010586977e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 260600, "loss": 0.5913056246936321, "lr": 0.0003, "grad_norm": 0.14564702367262997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38144860095, "tokens_per_second": 147791.88161933966, "data_wait_s": 1.672799771768041e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 260700, "loss": 0.6145157720893621, "lr": 0.0003, "grad_norm": 0.15188602167188195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38159524794, "tokens_per_second": 148843.88656316936, "data_wait_s": 9.1439978859853e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 260800, "loss": 0.5926761105656624, "lr": 0.0003, "grad_norm": 0.15779540370578413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38174072462, "tokens_per_second": 148441.49215839626, "data_wait_s": 1.4753997675143182e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a78b1698b21b044f38462d0bf0e73dce667e87f143a2b7eb66285d967829d732
-size 503928
+oid sha256:4b8c11d76f01305f3e684a4a3e055f714623ef939190e6c20c1e540c690b76ea
+size 506046