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": 76707.03690642263,
-      "required_pages_per_second_upper_estimate": 0.005306770798033466
+      "required_token_positions_per_second": 75960.15035579941,
+      "required_pages_per_second_upper_estimate": 0.005255099453445792
     },
     {
       "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": 9588.379613302828,
-      "required_pages_per_second_upper_estimate": 0.004014454315954868
+      "required_token_positions_per_second": 9495.018794474927,
+      "required_pages_per_second_upper_estimate": 0.003975366090707219
     },
     {
       "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": 19176.759226605656,
-      "required_pages_per_second_upper_estimate": 0.00399401322175261
+      "required_token_positions_per_second": 18990.037588949854,
+      "required_pages_per_second_upper_estimate": 0.003955124028809627
     },
     {
       "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": 28765.138839908486,
-      "required_pages_per_second_upper_estimate": 0.0017244261516286613
+      "required_token_positions_per_second": 28485.056383424777,
+      "required_pages_per_second_upper_estimate": 0.0017076356360236117
     },
     {
       "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": 28765.138839908486,
-      "required_pages_per_second_upper_estimate": 0.0022363630203286466
+      "required_token_positions_per_second": 28485.056383424777,
+      "required_pages_per_second_upper_estimate": 0.0022145878412895676
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28765.138839908486,
-      "required_pages_per_second_upper_estimate": 0.3689683154385973
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28485.056383424777,
+      "required_pages_per_second_upper_estimate": 0.36537571841593586
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14116032936721012,
-    "producer_p95_seconds": 0.3315256240020972,
-    "gpu_update_seconds": 0.9611634469721878,
-    "producer_batches_per_second": 7.084143289285128,
-    "gpu_batches_per_second": 1.0404057740128938,
-    "producer_headroom": 6.809019582774186,
+    "producer_batch_seconds": 0.2688127188597491,
+    "producer_p95_seconds": 0.34204886400402756,
+    "gpu_update_seconds": 0.9706141925029906,
+    "producer_batches_per_second": 3.7200620723669777,
+    "gpu_batches_per_second": 1.0302754768310467,
+    "producer_headroom": 3.610745044431476,
     "sustainable": true,
-    "buffer_seconds": 60.55329715924783,
+    "buffer_seconds": 60.17807993518542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 518469.02481020626,
-    "upload_bytes_per_second": 3876071.3599729273,
+    "generation_bytes_per_second": 485371.94957725826,
+    "upload_bytes_per_second": 3836117.136573954,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.25138338307993,
+    "estimated_full_shards_per_hour": 104.14952149856863,
     "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": 1.8011142590003146,
-      "bytes": 7499768,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.6402012090002245,
       "bytes": 10865294,
@@ -562,6 +556,12 @@
       "bytes": 5778338,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.6583789540018188,
+      "bytes": 2527741,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -362,3 +362,4 @@
 {"step": 836200, "loss": 0.5268641114234924, "lr": 3e-05, "grad_norm": 0.29490348842268344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122395894812, "tokens_per_second": 149424.52945566384, "data_wait_s": 1.2730000889860094e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 836300, "loss": 0.5561316013336182, "lr": 3e-05, "grad_norm": 0.2900443949403043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122410347896, "tokens_per_second": 149873.65379467962, "data_wait_s": 8.816001354716718e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 836400, "loss": 0.4622453451156616, "lr": 3e-05, "grad_norm": 0.295949444163284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122424908499, "tokens_per_second": 150424.64092202057, "data_wait_s": 8.505994628649205e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 836500, "loss": 0.5219582319259644, "lr": 3e-05, "grad_norm": 0.2973174038022154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122439594822, "tokens_per_second": 152322.13440774338, "data_wait_s": 8.138995326589793e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:408400eee8c2d46e5cb01a21d5839f175119a4e0d06b8381f79352b8472d1601
-size 776598
+oid sha256:1d74ee6905f38c5eefdf6de86c0b5f59ddc97023130f1255f489bb894593d101
+size 778716