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": 74340.36847476319,
-      "required_pages_per_second_upper_estimate": 0.0064436347254511386
+      "required_token_positions_per_second": 75209.5462473148,
+      "required_pages_per_second_upper_estimate": 0.006518972851864988
     },
     {
       "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": 9292.546059345399,
-      "required_pages_per_second_upper_estimate": 0.003706935559017632
+      "required_token_positions_per_second": 9401.19328091435,
+      "required_pages_per_second_upper_estimate": 0.003750276560122208
     },
     {
       "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": 18585.092118690798,
-      "required_pages_per_second_upper_estimate": 0.004299333975209217
+      "required_token_positions_per_second": 18802.3865618287,
+      "required_pages_per_second_upper_estimate": 0.004349601220377586
     },
     {
       "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": 27877.638178036195,
-      "required_pages_per_second_upper_estimate": 0.0016976019023786583
+      "required_token_positions_per_second": 28203.579842743053,
+      "required_pages_per_second_upper_estimate": 0.0017174500396755536
     },
     {
       "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": 27877.638178036195,
-      "required_pages_per_second_upper_estimate": 0.0021460977571868185
+      "required_token_positions_per_second": 28203.579842743053,
+      "required_pages_per_second_upper_estimate": 0.0021711896488002275
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27877.638178036195,
-      "required_pages_per_second_upper_estimate": 0.3461469657180699
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28203.579842743053,
+      "required_pages_per_second_upper_estimate": 0.35019407033715005
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11631141960165792,
-    "producer_p95_seconds": 0.2804522619990166,
-    "gpu_update_seconds": 0.991762638693793,
-    "producer_batches_per_second": 8.597608071716339,
-    "gpu_batches_per_second": 1.0083057790088323,
-    "producer_headroom": 8.526786467660449,
+    "producer_batch_seconds": 0.0847747782265742,
+    "producer_p95_seconds": 0.24619324099694495,
+    "gpu_update_seconds": 0.9803010878108083,
+    "producer_batches_per_second": 11.795961262527157,
+    "gpu_batches_per_second": 1.0200947570436578,
+    "producer_headroom": 11.563593657429529,
     "sustainable": true,
-    "buffer_seconds": 60.497520960321374,
+    "buffer_seconds": 60.77866744427011,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 267731.0046038409,
-    "upload_bytes_per_second": 5295046.569740084,
+    "generation_bytes_per_second": 267920.3541266785,
+    "upload_bytes_per_second": 5286393.6349620335,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.448841129173466,
+    "estimated_full_shards_per_hour": 57.489471128943116,
     "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": 1.1408444070002588,
-      "bytes": 5962047,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.150664095999673,
       "bytes": 11982825,
@@ -562,6 +556,12 @@
       "bytes": 6979691,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 1.2178401599994686,
+      "bytes": 5686805,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -191,3 +191,4 @@
 {"step": 256200, "loss": 0.6199516355991364, "lr": 0.0003, "grad_norm": 0.14335172639052332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37500863935, "tokens_per_second": 149267.98370844394, "data_wait_s": 1.6933001461438835e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256300, "loss": 0.5564228966832161, "lr": 0.0003, "grad_norm": 0.14486568315783913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37515491507, "tokens_per_second": 124273.96550886951, "data_wait_s": 2.0496001525316387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256400, "loss": 0.6124392151832581, "lr": 0.0003, "grad_norm": 0.1545684037659675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37530010083, "tokens_per_second": 147835.9191612433, "data_wait_s": 1.2881999282399192e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 256500, "loss": 0.5793410521000624, "lr": 0.0003, "grad_norm": 0.14140056791090877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37544598640, "tokens_per_second": 149039.37765120564, "data_wait_s": 1.3975000911159441e-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 sha256:422f0fed9e3d44e29a7f458718325426a63642c4f52d0bc36d7b3ac677edf27c
-size 412071
+oid sha256:589017199a52dcc6017a58bcfff1b5c2951f7bf8e9ee8148b6abe670ff7f54a4
+size 414189