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": 75903.59190667889,
-      "required_pages_per_second_upper_estimate": 0.005251186608701952
+      "required_token_positions_per_second": 76902.6533809988,
+      "required_pages_per_second_upper_estimate": 0.005320303999637402
     },
     {
       "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": 9487.948988334861,
-      "required_pages_per_second_upper_estimate": 0.003972406110510714
+      "required_token_positions_per_second": 9612.83167262485,
+      "required_pages_per_second_upper_estimate": 0.004024691882575936
     },
     {
       "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": 18975.897976669723,
-      "required_pages_per_second_upper_estimate": 0.0039521791204583274
+      "required_token_positions_per_second": 19225.6633452497,
+      "required_pages_per_second_upper_estimate": 0.004004198659977827
     },
     {
       "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": 28463.84696500458,
-      "required_pages_per_second_upper_estimate": 0.0017063641637742276
+      "required_token_positions_per_second": 28838.49501787455,
+      "required_pages_per_second_upper_estimate": 0.0017288237424893285
     },
     {
       "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": 28463.84696500458,
-      "required_pages_per_second_upper_estimate": 0.0022129389022977697
+      "required_token_positions_per_second": 28838.49501787455,
+      "required_pages_per_second_upper_estimate": 0.002242066140505781
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28463.84696500458,
-      "required_pages_per_second_upper_estimate": 0.36510366676934086
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28838.49501787455,
+      "required_pages_per_second_upper_estimate": 0.3699092497258187
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5730413844145801,
-    "producer_p95_seconds": 0.3138444790020003,
-    "gpu_update_seconds": 0.971337431443907,
-    "producer_batches_per_second": 1.7450746616173307,
-    "gpu_batches_per_second": 1.0295083537689735,
-    "producer_headroom": 1.6950563394932232,
+    "producer_batch_seconds": 0.2568817143204001,
+    "producer_p95_seconds": 0.3285221069963882,
+    "gpu_update_seconds": 0.9587185455712351,
+    "producer_batches_per_second": 3.892842285974209,
+    "gpu_batches_per_second": 1.0430589922553006,
+    "producer_headroom": 3.732140094547396,
     "sustainable": true,
-    "buffer_seconds": 60.22292074952224,
+    "buffer_seconds": 60.39926837098781,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283836.7356223343,
+    "generation_bytes_per_second": 281213.78190813365,
     "upload_bytes_per_second": 4929189.991331254,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.90475608351251,
+    "estimated_full_shards_per_hour": 60.34193127568252,
     "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": [

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

@@ -419,3 +419,4 @@
 {"step": 926900, "loss": 0.5539249181747437, "lr": 3e-05, "grad_norm": 0.3245176664790011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135671791690, "tokens_per_second": 153431.22632042432, "data_wait_s": 1.327700010733679e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927000, "loss": 0.5282608270645142, "lr": 3e-05, "grad_norm": 0.3187059187752664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135686554750, "tokens_per_second": 146545.1891244505, "data_wait_s": 1.5465004253201187e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927100, "loss": 0.5215455889701843, "lr": 3e-05, "grad_norm": 0.33811687711354393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135701278418, "tokens_per_second": 153998.26457812803, "data_wait_s": 1.2737000361084938e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 927200, "loss": 0.5180689692497253, "lr": 3e-05, "grad_norm": 0.3315590483140426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135715787627, "tokens_per_second": 151510.55934753237, "data_wait_s": 1.435800368199125e-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 sha256:5637d9bc44f8997c883bc00b42792918225e07932cb0dac57a1cdabc75bdbfa5
-size 898107
+oid sha256:cb4041ca31861642f7d6824f98ae047e568ad9ae019e84b983b341b04b3f64ff
+size 900225