Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77181.43590109641,
-      "required_pages_per_second_upper_estimate": 0.0053395908212423215
+      "required_token_positions_per_second": 77485.29949694258,
+      "required_pages_per_second_upper_estimate": 0.00536061280986364
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9647.679487637051,
-      "required_pages_per_second_upper_estimate": 0.004039281935016417
+      "required_token_positions_per_second": 9685.662437117822,
+      "required_pages_per_second_upper_estimate": 0.004055184602789836
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19295.358975274103,
-      "required_pages_per_second_upper_estimate": 0.004018714421714547
+      "required_token_positions_per_second": 19371.324874235645,
+      "required_pages_per_second_upper_estimate": 0.0040345361151127605
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28943.038462911154,
-      "required_pages_per_second_upper_estimate": 0.0017350909623906753
+      "required_token_positions_per_second": 29056.987311353467,
+      "required_pages_per_second_upper_estimate": 0.0017419220218649746
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28943.038462911154,
-      "required_pages_per_second_upper_estimate": 0.0022501939335193596
+      "required_token_positions_per_second": 29056.987311353467,
+      "required_pages_per_second_upper_estimate": 0.002259052955277734
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28943.038462911154,
-      "required_pages_per_second_upper_estimate": 0.3712502207887425
+      "required_token_positions_per_second": 29056.987311353467,
+      "required_pages_per_second_upper_estimate": 0.3727118342678194
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0733486928359639,
-    "producer_p95_seconds": 0.18050886999935756,
-    "gpu_update_seconds": 0.9552556147630916,
-    "producer_batches_per_second": 13.633508128582298,
-    "gpu_batches_per_second": 1.0468402221828397,
-    "producer_headroom": 13.02348518874649,
+    "producer_batch_seconds": 0.4187599488202878,
+    "producer_p95_seconds": 0.2293133739995028,
+    "gpu_update_seconds": 0.9515095183042968,
+    "producer_batches_per_second": 2.3880029664182456,
+    "gpu_batches_per_second": 1.050961635971986,
+    "producer_headroom": 2.272207552285857,
     "sustainable": true,
-    "buffer_seconds": 60.181103730074774,
+    "buffer_seconds": 60.89660917147499,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286379.026672346,
-    "upload_bytes_per_second": 5273935.104153618,
+    "generation_bytes_per_second": 285392.04199970147,
+    "upload_bytes_per_second": 5271899.522973409,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.45027256133828,
+    "estimated_full_shards_per_hour": 61.23848862641604,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 6531618,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1864704220006388,
+      "bytes": 6136727,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -69,3 +69,4 @@
 {"step": 891900, "loss": 0.5048050880432129, "lr": 3e-05, "grad_norm": 0.3236841189460622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130549141147, "tokens_per_second": 152273.73150246788, "data_wait_s": 9.749000128067564e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 892000, "loss": 0.5161802172660828, "lr": 3e-05, "grad_norm": 0.30836670864348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130563705809, "tokens_per_second": 152269.44595199273, "data_wait_s": 1.0167000255023595e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 892100, "loss": 0.502311110496521, "lr": 3e-05, "grad_norm": 0.31728557763669496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130578377600, "tokens_per_second": 153429.09644075166, "data_wait_s": 2.1569000637100544e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 892200, "loss": 0.48435771465301514, "lr": 3e-05, "grad_norm": 0.304577815562576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130592973653, "tokens_per_second": 152639.74397078506, "data_wait_s": 1.3949000276625156e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd1704625b827c4e841093e2408cba97cb874d544076f2220eb7bdc6228f959a
-size 151326
+oid sha256:2e5210edffda65a7e7556c7f4a4b0c413af8f3a2d41de2bb83fea0e2e52b9aaf
+size 153444