Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75296.87963433817,
-      "required_pages_per_second_upper_estimate": 0.006526542688507727
+      "required_token_positions_per_second": 75499.63362525223,
+      "required_pages_per_second_upper_estimate": 0.006544116890564866
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9412.109954292271,
-      "required_pages_per_second_upper_estimate": 0.0037546313843514726
+      "required_token_positions_per_second": 9437.454203156529,
+      "required_pages_per_second_upper_estimate": 0.0037647415841537135
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18824.219908584542,
-      "required_pages_per_second_upper_estimate": 0.00435465198089577
+      "required_token_positions_per_second": 18874.908406313058,
+      "required_pages_per_second_upper_estimate": 0.004366377872758171
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28236.32986287681,
-      "required_pages_per_second_upper_estimate": 0.0017194443440756231
+      "required_token_positions_per_second": 28312.362609469583,
+      "required_pages_per_second_upper_estimate": 0.0017240743394301313
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28236.32986287681,
-      "required_pages_per_second_upper_estimate": 0.002173710836007982
+      "required_token_positions_per_second": 28312.362609469583,
+      "required_pages_per_second_upper_estimate": 0.0021795640473128075
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28236.32986287681,
-      "required_pages_per_second_upper_estimate": 0.35060071597994474
+      "required_token_positions_per_second": 28312.362609469583,
+      "required_pages_per_second_upper_estimate": 0.35154478822739343
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4476953330153606,
-    "producer_p95_seconds": 0.3622250510015874,
-    "gpu_update_seconds": 0.9791640816730114,
-    "producer_batches_per_second": 2.233661881763886,
-    "gpu_batches_per_second": 1.021279291915394,
-    "producer_headroom": 2.187121485225346,
+    "producer_batch_seconds": 0.1345591942653641,
+    "producer_p95_seconds": 0.3552072899983614,
+    "gpu_update_seconds": 0.9765345401006069,
+    "producer_batches_per_second": 7.431673513352797,
+    "gpu_batches_per_second": 1.024029318918894,
+    "producer_headroom": 7.257285876539835,
     "sustainable": true,
-    "buffer_seconds": 60.708173063726704,
+    "buffer_seconds": 60.545141486237625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346451.82551959617,
+    "generation_bytes_per_second": 345220.7852088065,
     "upload_bytes_per_second": 4111165.837544698,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.3404967707721,
+    "estimated_full_shards_per_hour": 74.07634417722842,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -254,3 +254,4 @@
 {"step": 110400, "loss": 0.7208090014755726, "lr": 0.0003, "grad_norm": 0.17317937432345898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16154963944, "tokens_per_second": 148098.56580588644, "data_wait_s": 1.3700999261345714e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 110500, "loss": 0.606544328853488, "lr": 0.0003, "grad_norm": 0.1268284144823865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16169703709, "tokens_per_second": 149950.5660464123, "data_wait_s": 1.3818003935739398e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 110600, "loss": 0.6474052537232637, "lr": 0.0003, "grad_norm": 0.13868067379159402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16184359123, "tokens_per_second": 149237.21902385305, "data_wait_s": 1.58320035552606e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 110700, "loss": 0.6160413213074207, "lr": 0.0003, "grad_norm": 0.12751857205388983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16198989215, "tokens_per_second": 149054.01990013264, "data_wait_s": 1.3787997886538506e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4634bef003a62a3308d5f868be21a9c5ae8a59a074854e29fb64cef2324c9396
-size 546288
+oid sha256:02ea194d5009c53ca658f0d1b3f73fc8abe0d9ec2b5ced24e8749d4585ba6485
+size 548406