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": 76842.31754405703,
-      "required_pages_per_second_upper_estimate": 0.005316129826439341
+      "required_token_positions_per_second": 76079.62330939277,
+      "required_pages_per_second_upper_estimate": 0.005263364869590826
     },
     {
       "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": 9605.28969300713,
-      "required_pages_per_second_upper_estimate": 0.0040215342132044396
+      "required_token_positions_per_second": 9509.952913674097,
+      "required_pages_per_second_upper_estimate": 0.0039816186945560396
     },
     {
       "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": 19210.57938601426,
-      "required_pages_per_second_upper_estimate": 0.00400105706905984
+      "required_token_positions_per_second": 19019.905827348193,
+      "required_pages_per_second_upper_estimate": 0.0039613447951895855
     },
     {
       "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": 28815.869079021388,
-      "required_pages_per_second_upper_estimate": 0.0017274673520028898
+      "required_token_positions_per_second": 28529.85874102229,
+      "required_pages_per_second_upper_estimate": 0.0017103214689523434
     },
     {
       "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": 28815.869079021388,
-      "required_pages_per_second_upper_estimate": 0.002240307073280935
+      "required_token_positions_per_second": 28529.85874102229,
+      "required_pages_per_second_upper_estimate": 0.0022180710275280296
     },
     {
       "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": 28815.869079021388,
-      "required_pages_per_second_upper_estimate": 0.3696190284760507
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28529.85874102229,
+      "required_pages_per_second_upper_estimate": 0.3659503949541731
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07036065811724512,
-    "producer_p95_seconds": 0.16610220699931233,
-    "gpu_update_seconds": 0.9594713220059837,
-    "producer_batches_per_second": 14.212487869764594,
-    "gpu_batches_per_second": 1.0422406350919193,
-    "producer_headroom": 13.636474525397043,
+    "producer_batch_seconds": 0.3954433314219017,
+    "producer_p95_seconds": 0.16797798900006455,
+    "gpu_update_seconds": 0.9690899717020228,
+    "producer_batches_per_second": 2.5288073423928648,
+    "gpu_batches_per_second": 1.031895932473318,
+    "producer_headroom": 2.450641835879369,
     "sustainable": true,
-    "buffer_seconds": 60.44669328637698,
+    "buffer_seconds": 60.083578245525416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371684.31311117695,
-    "upload_bytes_per_second": 5004604.73724531,
+    "generation_bytes_per_second": 340682.87568087905,
+    "upload_bytes_per_second": 4994231.022069007,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.75480122567636,
+    "estimated_full_shards_per_hour": 73.1026144296625,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 1145768,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.216849121999985,
+      "bytes": 5739383,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -43,3 +43,4 @@
 {"step": 889300, "loss": 0.5247880816459656, "lr": 3e-05, "grad_norm": 0.30831380885092613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130169050578, "tokens_per_second": 152004.8896396374, "data_wait_s": 1.3217999367043376e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 889400, "loss": 0.5499499440193176, "lr": 3e-05, "grad_norm": 0.32768441044821467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130183539947, "tokens_per_second": 150971.01519701548, "data_wait_s": 1.8912000086857006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 889500, "loss": 0.5187528729438782, "lr": 3e-05, "grad_norm": 0.30286175324321074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130197951774, "tokens_per_second": 143902.7250845003, "data_wait_s": 2.0712999685201794e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 889600, "loss": 0.5349509716033936, "lr": 3e-05, "grad_norm": 0.33292206505396904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130212583858, "tokens_per_second": 152541.1211064751, "data_wait_s": 1.3189000128477346e-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 sha256:41a49eb04beee8d71949feeededf9476570ec0dd3be6376642922faf3fdf284d
-size 95475
+oid sha256:03336d29554f52d516b2f5246979196367ec153f1393cee06cb23dfc1ecb1f66
+size 97593