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": 73963.59467393602,
-      "required_pages_per_second_upper_estimate": 0.006410976927319907
+      "required_token_positions_per_second": 75518.28421008322,
+      "required_pages_per_second_upper_estimate": 0.006545733475988541
     },
     {
       "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": 9245.449334242003,
-      "required_pages_per_second_upper_estimate": 0.003688147971215096
+      "required_token_positions_per_second": 9439.785526260403,
+      "required_pages_per_second_upper_estimate": 0.0037656715837962353
     },
     {
       "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": 18490.898668484006,
-      "required_pages_per_second_upper_estimate": 0.004277543978251979
+      "required_token_positions_per_second": 18879.571052520805,
+      "required_pages_per_second_upper_estimate": 0.004367456493898563
     },
     {
       "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": 27736.34800272601,
-      "required_pages_per_second_upper_estimate": 0.0016889980719945277
+      "required_token_positions_per_second": 28319.356578781208,
+      "required_pages_per_second_upper_estimate": 0.0017245002354666882
     },
     {
       "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": 27736.34800272601,
-      "required_pages_per_second_upper_estimate": 0.0021352208483752023
+      "required_token_positions_per_second": 28319.356578781208,
+      "required_pages_per_second_upper_estimate": 0.0021801024624309626
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27736.34800272601,
-      "required_pages_per_second_upper_estimate": 0.34439261460851545
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28319.356578781208,
+      "required_pages_per_second_upper_estimate": 0.35163162991893426
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.3750802613594999,
-    "producer_p95_seconds": 0.32386904600207345,
-    "gpu_update_seconds": 0.9968147211479562,
-    "producer_batches_per_second": 2.6660960413524366,
-    "gpu_batches_per_second": 1.003195457274523,
-    "producer_headroom": 2.657603782014399,
+    "producer_batch_seconds": 0.08550038647646829,
+    "producer_p95_seconds": 0.31912344499869505,
+    "gpu_update_seconds": 0.9762933675094781,
+    "producer_batches_per_second": 11.695853565237666,
+    "gpu_batches_per_second": 1.0242822836654082,
+    "producer_headroom": 11.418584263103616,
     "sustainable": true,
-    "buffer_seconds": 60.80569799002533,
+    "buffer_seconds": 60.53018878558765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343033.563595588,
-    "upload_bytes_per_second": 4986447.894074069,
+    "generation_bytes_per_second": 341951.3222658888,
+    "upload_bytes_per_second": 5089541.067409014,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.60701733494501,
+    "estimated_full_shards_per_hour": 73.37479353888033,
     "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": 0.8186183350007923,
-      "bytes": 971258,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.5642075370014936,
       "bytes": 10184682,
@@ -562,6 +556,12 @@
       "bytes": 6821427,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.3006846259995655,
+      "bytes": 12006518,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -315,3 +315,4 @@
 {"step": 386500, "loss": 0.5747383907437325, "lr": 0.0003, "grad_norm": 0.16812191695093157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56572755382, "tokens_per_second": 149876.05576409269, "data_wait_s": 9.35700154514052e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 386600, "loss": 0.6133938077837229, "lr": 0.0003, "grad_norm": 0.15190493584300963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56587505244, "tokens_per_second": 144101.1346329765, "data_wait_s": 8.521001291228458e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 386700, "loss": 0.6168648153543472, "lr": 0.0003, "grad_norm": 0.17218091571528626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56602285564, "tokens_per_second": 151670.51335805992, "data_wait_s": 9.251998562831432e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 386800, "loss": 0.544911116361618, "lr": 0.0003, "grad_norm": 0.1376656132044407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56616993030, "tokens_per_second": 149859.28636146747, "data_wait_s": 1.2904998584417626e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:222af66e0bff9266c6052ae5d3f987362dbc5b92fcb5bc6761bbe63c8b1887f5
-size 676269
+oid sha256:d414539875ecd925c356adabb0487cb00dc73689fa6cf5baa243bbca2340722a
+size 678387