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": 75493.97147400327,
-      "required_pages_per_second_upper_estimate": 0.006543626109645197
+      "required_token_positions_per_second": 73895.66897117315,
+      "required_pages_per_second_upper_estimate": 0.0064050892995605405
     },
     {
       "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": 9436.746434250408,
-      "required_pages_per_second_upper_estimate": 0.003764459244554974
+      "required_token_positions_per_second": 9236.958621396643,
+      "required_pages_per_second_upper_estimate": 0.0036847608989136124
     },
     {
       "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": 18873.492868500816,
-      "required_pages_per_second_upper_estimate": 0.004366050412997924
+      "required_token_positions_per_second": 18473.917242793286,
+      "required_pages_per_second_upper_estimate": 0.0042736156242813165
     },
     {
       "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": 28310.239302751223,
-      "required_pages_per_second_upper_estimate": 0.0017239450411911125
+      "required_token_positions_per_second": 27710.87586418993,
+      "required_pages_per_second_upper_estimate": 0.001687446952399663
     },
     {
       "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": 28310.239302751223,
-      "required_pages_per_second_upper_estimate": 0.0021794005892839387
+      "required_token_positions_per_second": 27710.87586418993,
+      "required_pages_per_second_upper_estimate": 0.002133259932639303
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28310.239302751223,
-      "required_pages_per_second_upper_estimate": 0.35151842386420185
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27710.87586418993,
+      "required_pages_per_second_upper_estimate": 0.34407633589766107
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.3600402284452855,
-    "producer_p95_seconds": 0.3075554220013146,
-    "gpu_update_seconds": 0.976607781528471,
-    "producer_batches_per_second": 2.777467407789871,
-    "gpu_batches_per_second": 1.0239525210775182,
-    "producer_headroom": 2.712496283389299,
+    "producer_batch_seconds": 0.2676466055233675,
+    "producer_p95_seconds": 0.34545705399796134,
+    "gpu_update_seconds": 0.997731004083087,
+    "producer_batches_per_second": 3.736270064193632,
+    "gpu_batches_per_second": 1.0022741559675177,
+    "producer_headroom": 3.7277924826734923,
     "sustainable": true,
-    "buffer_seconds": 60.549682454765204,
+    "buffer_seconds": 60.861591249068304,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390084.24814003514,
-    "upload_bytes_per_second": 5477015.451686867,
+    "generation_bytes_per_second": 387110.9417306228,
+    "upload_bytes_per_second": 5516682.817063272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.70299895430365,
+    "estimated_full_shards_per_hour": 83.06499661387456,
     "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": 0.9494441940005345,
-      "bytes": 2648505,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0081078959992738,
       "bytes": 5874669,
@@ -562,6 +556,12 @@
       "bytes": 4770134,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0199386870008311,
+      "bytes": 6036267,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -203,3 +203,4 @@
 {"step": 335300, "loss": 0.5784935504198074, "lr": 0.0003, "grad_norm": 0.2093674290392367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49079527448, "tokens_per_second": 148808.59507893154, "data_wait_s": 9.47700027609244e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 335400, "loss": 0.600412305444479, "lr": 0.0003, "grad_norm": 0.1508263100708454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49094194312, "tokens_per_second": 148633.22685352326, "data_wait_s": 9.461997251491994e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 335500, "loss": 0.5917552970349789, "lr": 0.0003, "grad_norm": 0.14277662475239128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49108894350, "tokens_per_second": 148414.07101879962, "data_wait_s": 1.9260001863585785e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 335600, "loss": 0.5991566870361567, "lr": 0.0003, "grad_norm": 0.16026576510517238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49123424415, "tokens_per_second": 147370.17736312214, "data_wait_s": 1.369399979012087e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a724762237c873341c237df9240ce74fa6d20f797eb61e8650d7d3e372ab6c31
-size 437487
+oid sha256:255f003a2309360df6700578cba164839a235a96a55dc4eeedf7b4bd623e4e2a
+size 439605