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": 77222.58994479511,
-      "required_pages_per_second_upper_estimate": 0.005342437953475935
+      "required_token_positions_per_second": 75631.67090441682,
+      "required_pages_per_second_upper_estimate": 0.00523237448282181
     },
     {
       "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": 9652.823743099389,
-      "required_pages_per_second_upper_estimate": 0.0040414357273542275
+      "required_token_positions_per_second": 9453.958863052103,
+      "required_pages_per_second_upper_estimate": 0.0039581751548493525
     },
     {
       "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": 19305.647486198777,
-      "required_pages_per_second_upper_estimate": 0.004020857247213877
+      "required_token_positions_per_second": 18907.917726104206,
+      "required_pages_per_second_upper_estimate": 0.003938020627025296
     },
     {
       "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": 28958.471229298164,
-      "required_pages_per_second_upper_estimate": 0.0017360161331711067
+      "required_token_positions_per_second": 28361.87658915631,
+      "required_pages_per_second_upper_estimate": 0.0017002511954418724
     },
     {
       "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": 28958.471229298164,
-      "required_pages_per_second_upper_estimate": 0.0022513937632243843
+      "required_token_positions_per_second": 28361.87658915631,
+      "required_pages_per_second_upper_estimate": 0.0022050111540958487
     },
     {
       "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": 28958.471229298164,
-      "required_pages_per_second_upper_estimate": 0.37144817574554156
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28361.87658915631,
+      "required_pages_per_second_upper_estimate": 0.3637957002752185
     }
   ],
   "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": 0.31641568100758377,
-    "producer_p95_seconds": 0.3251587750055478,
-    "gpu_update_seconds": 0.9547465327530025,
-    "producer_batches_per_second": 3.1603996262626195,
-    "gpu_batches_per_second": 1.0473984096244995,
-    "producer_headroom": 3.017380585288121,
+    "producer_batch_seconds": 0.09063524006239732,
+    "producer_p95_seconds": 0.32995927400043,
+    "gpu_update_seconds": 0.974829712451776,
+    "producer_batches_per_second": 11.033236071439273,
+    "gpu_batches_per_second": 1.0258201891332577,
+    "producer_headroom": 10.75552634693371,
     "sustainable": true,
-    "buffer_seconds": 60.14903156343916,
+    "buffer_seconds": 60.43944217201011,
     "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": 364633.51407461765,
-    "upload_bytes_per_second": 5657426.298353266,
+    "generation_bytes_per_second": 364533.2995165233,
+    "upload_bytes_per_second": 5619729.502013892,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.24186388663193,
+    "estimated_full_shards_per_hour": 78.22036017534042,
     "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": 1.2100134370048181,
-      "bytes": 12201541,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.149907068000175,
       "bytes": 11779236,
@@ -562,6 +556,12 @@
       "bytes": 2735426,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4271837970009074,
+      "bytes": 10546211,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -388,3 +388,4 @@
 {"step": 723800, "loss": 0.5341401100158691, "lr": 8e-05, "grad_norm": 0.227167872318649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105944292649, "tokens_per_second": 152246.5032538068, "data_wait_s": 2.308800321770832e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723900, "loss": 0.5106531381607056, "lr": 8e-05, "grad_norm": 0.22065503390254373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105958933471, "tokens_per_second": 152128.62594890653, "data_wait_s": 1.9661994883790612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724000, "loss": 0.5931006073951721, "lr": 8e-05, "grad_norm": 0.2112529032189782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105973600123, "tokens_per_second": 151937.48000324584, "data_wait_s": 1.2025004252791405e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 724100, "loss": 0.5540884733200073, "lr": 8e-05, "grad_norm": 0.22176082356154947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105988236583, "tokens_per_second": 152912.49624472822, "data_wait_s": 1.2636999599635601e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3a3bfe352ba1ce7cd70d018d2d48a2149ccb252fe81a30ec055360f92ca1234
-size 831666
+oid sha256:9409538b8b30f2c26ec9cb4a40a4e37b42704d784530ee1481faa1789d69f57e
+size 833784