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": 75205.37167677315,
-      "required_pages_per_second_upper_estimate": 0.0065186110106707894
+      "required_token_positions_per_second": 74927.0428729743,
+      "required_pages_per_second_upper_estimate": 0.0064944861753754085
     },
     {
       "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": 9400.671459596644,
-      "required_pages_per_second_upper_estimate": 0.0037500683977966507
+      "required_token_positions_per_second": 9365.880359121787,
+      "required_pages_per_second_upper_estimate": 0.0037361897076439235
     },
     {
       "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": 18801.34291919329,
-      "required_pages_per_second_upper_estimate": 0.004349359792021366
+      "required_token_positions_per_second": 18731.760718243575,
+      "required_pages_per_second_upper_estimate": 0.004333263174436561
     },
     {
       "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": 28202.014378789932,
-      "required_pages_per_second_upper_estimate": 0.0017173547111341632
+      "required_token_positions_per_second": 28097.641077365362,
+      "required_pages_per_second_upper_estimate": 0.0017109989246817982
     },
     {
       "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": 28202.014378789932,
-      "required_pages_per_second_upper_estimate": 0.0021710691350516347
+      "required_token_positions_per_second": 28097.641077365362,
+      "required_pages_per_second_upper_estimate": 0.002163034189384181
     },
     {
       "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": 28202.014378789932,
-      "required_pages_per_second_upper_estimate": 0.3501746325141231
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28097.641077365362,
+      "required_pages_per_second_upper_estimate": 0.348878665425399
     }
   ],
   "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": 0.14947177726486416,
-    "producer_p95_seconds": 0.4030763749979087,
-    "gpu_update_seconds": 0.9803555032860847,
-    "producer_batches_per_second": 6.690226197203763,
-    "gpu_batches_per_second": 1.0200381358069275,
-    "producer_headroom": 6.558800070657443,
+    "producer_batch_seconds": 0.11899730081228199,
+    "producer_p95_seconds": 0.40818131800187984,
+    "gpu_update_seconds": 0.9839971947777645,
+    "producer_batches_per_second": 8.40355195600191,
+    "gpu_batches_per_second": 1.016263059800541,
+    "producer_headroom": 8.269071550875076,
     "sustainable": true,
-    "buffer_seconds": 60.78204120373725,
+    "buffer_seconds": 60.023828881443634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358088.58944669,
-    "upload_bytes_per_second": 5728882.462079497,
+    "generation_bytes_per_second": 396198.53554834897,
+    "upload_bytes_per_second": 5722493.002090204,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.83747541952634,
+    "estimated_full_shards_per_hour": 85.01498269880153,
     "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": 1.0975180309978896,
-      "bytes": 7653776,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9772678519984765,
       "bytes": 5704498,
@@ -562,6 +556,12 @@
       "bytes": 846588,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1457775259987102,
+      "bytes": 7463584,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -372,3 +372,4 @@
 {"step": 232200, "loss": 0.5945616010576487, "lr": 0.0003, "grad_norm": 0.1442751482313228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33987578625, "tokens_per_second": 146378.1621474991, "data_wait_s": 1.733299723127857e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232300, "loss": 0.6218109801411629, "lr": 0.0003, "grad_norm": 0.1415363212380993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34002106053, "tokens_per_second": 148234.62647555297, "data_wait_s": 1.3174998457543552e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232400, "loss": 0.6167115028947592, "lr": 0.0003, "grad_norm": 0.1588070864344922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34016744106, "tokens_per_second": 148924.43527422822, "data_wait_s": 1.2810996850021183e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 232500, "loss": 0.5852804183959961, "lr": 0.0003, "grad_norm": 0.15621871718730998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34031515048, "tokens_per_second": 150568.59915837742, "data_wait_s": 9.556999430060387e-06, "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:afeea82391559ecba7f755295bc23203e4a4979b50ccd083c04d4208a95029cf
-size 797778
+oid sha256:c5076f1f89e8d9ffc2bfce75206e2ded8cdde6ba8b5f800303efad3056245d19
+size 799896