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": 76835.46975774648,
-      "required_pages_per_second_upper_estimate": 0.005315656080693323
+      "required_token_positions_per_second": 76094.9215805354,
+      "required_pages_per_second_upper_estimate": 0.005264423239485353
     },
     {
       "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": 9604.43371971831,
-      "required_pages_per_second_upper_estimate": 0.004021175835063166
+      "required_token_positions_per_second": 9511.865197566925,
+      "required_pages_per_second_upper_estimate": 0.003982419327888938
     },
     {
       "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": 19208.86743943662,
-      "required_pages_per_second_upper_estimate": 0.004000700515734785
+      "required_token_positions_per_second": 19023.73039513385,
+      "required_pages_per_second_upper_estimate": 0.003962141351798704
     },
     {
       "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": 28813.30115915493,
-      "required_pages_per_second_upper_estimate": 0.0017273134091278822
+      "required_token_positions_per_second": 28535.595592700778,
+      "required_pages_per_second_upper_estimate": 0.0017106653844508035
     },
     {
       "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": 28813.30115915493,
-      "required_pages_per_second_upper_estimate": 0.0022401074288064034
+      "required_token_positions_per_second": 28535.595592700778,
+      "required_pages_per_second_upper_estimate": 0.002218517042512288
     },
     {
       "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": 28813.30115915493,
-      "required_pages_per_second_upper_estimate": 0.3695860899572213
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28535.595592700778,
+      "required_pages_per_second_upper_estimate": 0.36602398112775336
     }
   ],
   "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": 0.1519998379062315,
-    "producer_p95_seconds": 0.21432510100203217,
-    "gpu_update_seconds": 0.9595568327031255,
-    "producer_batches_per_second": 6.578954384259927,
-    "gpu_batches_per_second": 1.0421477560458237,
-    "producer_headroom": 6.312880631458797,
+    "producer_batch_seconds": 0.11536133788280267,
+    "producer_p95_seconds": 0.21994551100215176,
+    "gpu_update_seconds": 0.9688951439679143,
+    "producer_batches_per_second": 8.668415418482015,
+    "gpu_batches_per_second": 1.032103428555439,
+    "producer_headroom": 8.39878560486382,
     "sustainable": true,
-    "buffer_seconds": 60.452080460296905,
+    "buffer_seconds": 60.07149892601069,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392167.2770069212,
-    "upload_bytes_per_second": 3680890.701886359,
+    "generation_bytes_per_second": 393554.6069193899,
+    "upload_bytes_per_second": 3671954.09817317,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.14996845870711,
+    "estimated_full_shards_per_hour": 84.4476571625354,
     "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": 1.5349632739998924,
-      "bytes": 8020716,
-      "pages": 48,
-      "shards": 3
-    },
     {
       "seconds": 1.4110580969991133,
       "bytes": 5973740,
@@ -562,6 +556,12 @@
       "bytes": 10660170,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.4908248419997108,
+      "bytes": 6812358,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 941900, "loss": 0.49831411242485046, "lr": 3e-05, "grad_norm": 0.3497331215042466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137867413127, "tokens_per_second": 152633.72317608906, "data_wait_s": 1.2627999240066856e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942000, "loss": 0.5415322184562683, "lr": 3e-05, "grad_norm": 0.3412821098691674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137881971065, "tokens_per_second": 151928.18644188144, "data_wait_s": 1.5634002920705825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942100, "loss": 0.5240259766578674, "lr": 3e-05, "grad_norm": 0.3313426137066382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137896578923, "tokens_per_second": 151914.21895016293, "data_wait_s": 1.0146999557036906e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 942200, "loss": 0.5340875387191772, "lr": 3e-05, "grad_norm": 0.3456524713919745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137911204235, "tokens_per_second": 152384.73326796383, "data_wait_s": 1.0294999810867012e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a950236f3597cd30969126469dd75e587de6c9895164e57b18e849125c70e712
-size 364692
+oid sha256:347ee0f1ebdc90228da7666608f187a8dbff4976f6411c94aec558fa1b5c9507
+size 366810