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": 75773.9273110939,
-      "required_pages_per_second_upper_estimate": 0.00656789197206263
+      "required_token_positions_per_second": 75167.43451422501,
+      "required_pages_per_second_upper_estimate": 0.006515322713571975
     },
     {
       "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": 9471.740913886737,
-      "required_pages_per_second_upper_estimate": 0.0037784190656960015
+      "required_token_positions_per_second": 9395.929314278126,
+      "required_pages_per_second_upper_estimate": 0.003748176685127703
     },
     {
       "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": 18943.481827773474,
-      "required_pages_per_second_upper_estimate": 0.00438224112696204
+      "required_token_positions_per_second": 18791.858628556252,
+      "required_pages_per_second_upper_estimate": 0.004347165768300302
     },
     {
       "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": 28415.222741660215,
-      "required_pages_per_second_upper_estimate": 0.0017303379818151346
+      "required_token_positions_per_second": 28187.787942834377,
+      "required_pages_per_second_upper_estimate": 0.0017164883958248108
     },
     {
       "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": 28415.222741660215,
-      "required_pages_per_second_upper_estimate": 0.002187482505023912
+      "required_token_positions_per_second": 28187.787942834377,
+      "required_pages_per_second_upper_estimate": 0.002169973944630481
     },
     {
       "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": 28415.222741660215,
-      "required_pages_per_second_upper_estimate": 0.3528219668184836
+      "required_token_positions_per_second": 28187.787942834377,
+      "required_pages_per_second_upper_estimate": 0.3499979877923734
     }
   ],
   "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": 0.09463657644550949,
-    "producer_p95_seconds": 0.37145910199615173,
-    "gpu_update_seconds": 0.9729995872763169,
-    "producer_batches_per_second": 10.566738966680468,
-    "gpu_batches_per_second": 1.0277496651352798,
-    "producer_headroom": 10.281432653436669,
+    "producer_batch_seconds": 0.13924695535934006,
+    "producer_p95_seconds": 0.36937078799383016,
+    "gpu_update_seconds": 0.9808502907738244,
+    "producer_batches_per_second": 7.181485565838079,
+    "gpu_batches_per_second": 1.0195235801083036,
+    "producer_headroom": 7.043962205440303,
     "sustainable": true,
-    "buffer_seconds": 60.32597441113165,
+    "buffer_seconds": 60.81271802797711,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 160,
+    "upload_pages_per_commit": 151,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 604102.887834581,
+    "generation_bytes_per_second": 618620.3208053231,
     "upload_bytes_per_second": 5614849.837062628,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.81320846690213,
+    "estimated_full_shards_per_hour": 66.37076005039106,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 228200, "loss": 0.5915424004197121, "lr": 0.0003, "grad_norm": 0.15198145500474589, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33401767184, "tokens_per_second": 147298.9726504159, "data_wait_s": 1.1588002962525934e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 228300, "loss": 0.6020775269716978, "lr": 0.0003, "grad_norm": 0.15287950696838631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33416375290, "tokens_per_second": 148827.58254269732, "data_wait_s": 1.4884004485793412e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 228400, "loss": 0.5106710838153958, "lr": 0.0003, "grad_norm": 0.14505354656055472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33431002927, "tokens_per_second": 149035.01883678947, "data_wait_s": 1.1543997970875353e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 228500, "loss": 0.6087172422558069, "lr": 0.0003, "grad_norm": 0.1542153343676074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33445805351, "tokens_per_second": 150633.85695596994, "data_wait_s": 9.520997991785407e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:531aa8ddf0543ec509d13c4ec3ddac29764ac51bab10ba90b1beccd40d1ab62b
-size 712275
+oid sha256:da1febf7ae838acadf102af9e34d0b14cfb11739ad8eb5d5dec754041fa3e2ab
+size 714393