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": 74294.54794814497,
-      "required_pages_per_second_upper_estimate": 0.006439663118334934
+      "required_token_positions_per_second": 73422.34216076705,
+      "required_pages_per_second_upper_estimate": 0.006364062531270373
     },
     {
       "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": 9286.818493518122,
-      "required_pages_per_second_upper_estimate": 0.00370465074737439
+      "required_token_positions_per_second": 9177.79277009588,
+      "required_pages_per_second_upper_estimate": 0.00366115875622143
     },
     {
       "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": 18573.636987036243,
-      "required_pages_per_second_upper_estimate": 0.004296684032104367
+      "required_token_positions_per_second": 18355.58554019176,
+      "required_pages_per_second_upper_estimate": 0.004246241667451291
     },
     {
       "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": 27860.455480554367,
-      "required_pages_per_second_upper_estimate": 0.0016965555662526615
+      "required_token_positions_per_second": 27533.37831028764,
+      "required_pages_per_second_upper_estimate": 0.0016766382826247075
     },
     {
       "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": 27860.455480554367,
-      "required_pages_per_second_upper_estimate": 0.0021447749855699164
+      "required_token_positions_per_second": 27533.37831028764,
+      "required_pages_per_second_upper_estimate": 0.002119595679595228
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27860.455480554367,
-      "required_pages_per_second_upper_estimate": 0.34593361412213475
+      "required_token_positions_per_second": 27533.37831028764,
+      "required_pages_per_second_upper_estimate": 0.34187241032429366
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.34345765746894585,
-    "producer_p95_seconds": 0.42753673200058984,
-    "gpu_update_seconds": 0.9923742998134882,
-    "producer_batches_per_second": 2.9115670542020635,
-    "gpu_batches_per_second": 1.0076842983418102,
-    "producer_headroom": 2.889364316773793,
+    "producer_batch_seconds": 0.12484272595315815,
+    "producer_p95_seconds": 0.39997743900312344,
+    "gpu_update_seconds": 1.0041630085643916,
+    "producer_batches_per_second": 8.010078219336599,
+    "gpu_batches_per_second": 0.9958542502274176,
+    "producer_headroom": 8.043424243565143,
     "sustainable": true,
-    "buffer_seconds": 60.53483228862278,
+    "buffer_seconds": 60.24978051386349,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435649.7104628073,
-    "upload_bytes_per_second": 5640056.339896252,
+    "generation_bytes_per_second": 437238.3343946642,
+    "upload_bytes_per_second": 5718900.184719799,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.48028645909467,
+    "estimated_full_shards_per_hour": 93.82116817359872,
     "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": 1.1756158520001918,
-      "bytes": 5451054,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.2875092799986305,
       "bytes": 13122964,
@@ -562,6 +556,12 @@
       "bytes": 2591189,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2582520649957587,
+      "bytes": 11695145,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 230300, "loss": 0.6145809032022953, "lr": 0.0003, "grad_norm": 0.14974299864634225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33709652932, "tokens_per_second": 148762.6084711953, "data_wait_s": 1.1199001164641231e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 230400, "loss": 0.6153337731957436, "lr": 0.0003, "grad_norm": 0.1430661918853785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33724108828, "tokens_per_second": 146262.5002708152, "data_wait_s": 1.1616000847425312e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 230500, "loss": 0.5772615559399128, "lr": 0.0003, "grad_norm": 0.17446808019470036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33738817563, "tokens_per_second": 149107.7870497439, "data_wait_s": 1.2322001566644758e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 230600, "loss": 0.5562775582075119, "lr": 0.0003, "grad_norm": 0.15923243384969488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33753458501, "tokens_per_second": 148866.19270517267, "data_wait_s": 1.694400270935148e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a2fb1075c93f3361b2485367841491fb692ab8e40d1e4bfa29b491e10f1d815
-size 757536
+oid sha256:5cde7d9e8c49cd4b94d13078324e21746c0123779f634e8f73ad2d3046fb7e00
+size 759654