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": 75205.36431719418,
-      "required_pages_per_second_upper_estimate": 0.006518610372761136
+      "required_token_positions_per_second": 74463.23807554158,
+      "required_pages_per_second_upper_estimate": 0.006454284751036453
     },
     {
       "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": 9400.670539649273,
-      "required_pages_per_second_upper_estimate": 0.00375006803081589
+      "required_token_positions_per_second": 9307.904759442697,
+      "required_pages_per_second_upper_estimate": 0.003713062374119474
     },
     {
       "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": 18801.341079298545,
-      "required_pages_per_second_upper_estimate": 0.004349359366394098
+      "required_token_positions_per_second": 18615.809518885395,
+      "required_pages_per_second_upper_estimate": 0.0043064399051364576
     },
     {
       "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": 28202.01161894782,
-      "required_pages_per_second_upper_estimate": 0.0017173545430742582
+      "required_token_positions_per_second": 27923.71427832809,
+      "required_pages_per_second_upper_estimate": 0.0017004076951438194
     },
     {
       "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": 28202.01161894782,
-      "required_pages_per_second_upper_estimate": 0.002171068922591349
+      "required_token_positions_per_second": 27923.71427832809,
+      "required_pages_per_second_upper_estimate": 0.002149644822934097
     },
     {
       "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": 28202.01161894782,
-      "required_pages_per_second_upper_estimate": 0.3501745982461207
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27923.71427832809,
+      "required_pages_per_second_upper_estimate": 0.3467190766769074
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11174089453906788,
-    "producer_p95_seconds": 0.21262714200020127,
-    "gpu_update_seconds": 0.9803555992234397,
-    "producer_batches_per_second": 8.949275053909389,
-    "gpu_batches_per_second": 1.0200380359862493,
-    "producer_headroom": 8.77347190809072,
+    "producer_batch_seconds": 0.0931099763124621,
+    "producer_p95_seconds": 0.25880407499971625,
+    "gpu_update_seconds": 0.9901261603101964,
+    "producer_batches_per_second": 10.739987696314742,
+    "gpu_batches_per_second": 1.0099723046270288,
+    "producer_headroom": 10.633942779530868,
     "sustainable": true,
-    "buffer_seconds": 60.78204715185326,
+    "buffer_seconds": 60.39769577892198,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491188.2842349423,
-    "upload_bytes_per_second": 5194652.380403602,
+    "generation_bytes_per_second": 493945.63880467275,
+    "upload_bytes_per_second": 5262319.0625148835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.39757151876641,
+    "estimated_full_shards_per_hour": 105.98923562150132,
     "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": 0.9170148619996326,
-      "bytes": 257577,
-      "pages": 8,
-      "shards": 1
-    },
     {
       "seconds": 1.4457078639998144,
       "bytes": 7654224,
@@ -562,6 +556,12 @@
       "bytes": 2318063,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9941544879984576,
+      "bytes": 6059545,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -126,3 +126,4 @@
 {"step": 207600, "loss": 0.569626696407795, "lr": 0.0003, "grad_norm": 0.14743573959776018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30384928391, "tokens_per_second": 147426.8888508114, "data_wait_s": 1.8900000213761814e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207700, "loss": 0.5794225446879864, "lr": 0.0003, "grad_norm": 0.14461750445336438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30399624366, "tokens_per_second": 148501.17824133413, "data_wait_s": 1.809899913496338e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207800, "loss": 0.5990188978612423, "lr": 0.0003, "grad_norm": 0.15050431338140977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30414300144, "tokens_per_second": 147051.9778778555, "data_wait_s": 1.845699989644345e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 207900, "loss": 0.6236999053508043, "lr": 0.0003, "grad_norm": 0.14992720926075997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30428967788, "tokens_per_second": 145045.0561079477, "data_wait_s": 1.9193001207895577e-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 sha256:5e124cb8c4e003f30502cc2b27c9c2754b38b80bd24efeacfb28c4ef634245a5
-size 272835
+oid sha256:67b94de47ba526a91fdef6b277106741d708161b4ee72a21ef7884683beaca35
+size 274953