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": 73838.9738571702,
-      "required_pages_per_second_upper_estimate": 0.006400175110771218
+      "required_token_positions_per_second": 74105.52193015984,
+      "required_pages_per_second_upper_estimate": 0.006423278822178045
     },
     {
       "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": 9229.871732146275,
-      "required_pages_per_second_upper_estimate": 0.003681933832833204
+      "required_token_positions_per_second": 9263.19024126998,
+      "required_pages_per_second_upper_estimate": 0.003695225084278754
     },
     {
       "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": 18459.74346429255,
-      "required_pages_per_second_upper_estimate": 0.004270336770075153
+      "required_token_positions_per_second": 18526.38048253996,
+      "required_pages_per_second_upper_estimate": 0.004285752071475225
     },
     {
       "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": 27689.615196438826,
-      "required_pages_per_second_upper_estimate": 0.0016861522892797239
+      "required_token_positions_per_second": 27789.570723809942,
+      "required_pages_per_second_upper_estimate": 0.0016922390564705024
     },
     {
       "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": 27689.615196438826,
-      "required_pages_per_second_upper_estimate": 0.0021316232275825275
+      "required_token_positions_per_second": 27789.570723809942,
+      "required_pages_per_second_upper_estimate": 0.002139318080774165
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27689.615196438826,
-      "required_pages_per_second_upper_estimate": 0.3438123495590701
+      "required_token_positions_per_second": 27789.570723809942,
+      "required_pages_per_second_upper_estimate": 0.34505346267938886
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06087142204689222,
-    "producer_p95_seconds": 0.131672928999933,
-    "gpu_update_seconds": 0.9984970828903329,
-    "producer_batches_per_second": 16.42807028936586,
-    "gpu_batches_per_second": 1.0015051792693441,
-    "producer_headroom": 16.403380261449158,
+    "producer_batch_seconds": 0.07913221451562791,
+    "producer_p95_seconds": 0.14340419199993448,
+    "gpu_update_seconds": 0.9949056167431675,
+    "producer_batches_per_second": 12.63707841516945,
+    "gpu_batches_per_second": 1.0051204688878017,
+    "producer_headroom": 12.572700294475931,
     "sustainable": true,
-    "buffer_seconds": 60.908322056310304,
+    "buffer_seconds": 60.689242621333214,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 71,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472803.54582325416,
+    "generation_bytes_per_second": 466573.9417329422,
     "upload_bytes_per_second": 4695517.6851320835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.45263463042468,
+    "estimated_full_shards_per_hour": 100.11590661040496,
     "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-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -9,3 +9,4 @@
 {"step": 82800, "loss": 0.5665384028106928, "lr": 0.0003, "grad_norm": 0.13098018386505245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12117034475, "tokens_per_second": 145526.04640431632, "data_wait_s": 1.0684000017135986e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 82900, "loss": 0.5804585367441177, "lr": 0.0003, "grad_norm": 0.13427587039895614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12131681070, "tokens_per_second": 149314.93150573867, "data_wait_s": 1.162500007012568e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 83000, "loss": 0.6183635629713535, "lr": 0.0003, "grad_norm": 0.13188186813879005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12146491385, "tokens_per_second": 150376.761259858, "data_wait_s": 1.3017000128456857e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 83100, "loss": 0.6229391451925039, "lr": 0.0003, "grad_norm": 0.13720698418826618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12160969657, "tokens_per_second": 146814.05009065697, "data_wait_s": 1.158499981102068e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91787c870d5a21501d741eb630bc0722ed0a88ad11b504ef49740a989e8ff96b
-size 23463
+oid sha256:92982570bcd02057ad2ef0ec6db8b30d2144e7ac60651b30c76efb4cefc835a3
+size 25581