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": 75400.62889578588,
-      "required_pages_per_second_upper_estimate": 0.006535535411539915
+      "required_token_positions_per_second": 74917.47006177217,
+      "required_pages_per_second_upper_estimate": 0.006493656428362476
     },
     {
       "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": 9425.078611973235,
-      "required_pages_per_second_upper_estimate": 0.00375980477579912
+      "required_token_positions_per_second": 9364.683757721521,
+      "required_pages_per_second_upper_estimate": 0.003735712365454572
     },
     {
       "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": 18850.15722394647,
-      "required_pages_per_second_upper_estimate": 0.004360652122323593
+      "required_token_positions_per_second": 18729.367515443042,
+      "required_pages_per_second_upper_estimate": 0.0043327095490875885
     },
     {
       "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": 28275.235835919706,
-      "required_pages_per_second_upper_estimate": 0.0017218135137100706
+      "required_token_positions_per_second": 28094.05127316456,
+      "required_pages_per_second_upper_estimate": 0.0017107803244402165
     },
     {
       "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": 28275.235835919706,
-      "required_pages_per_second_upper_estimate": 0.002176705925511451
+      "required_token_positions_per_second": 28094.05127316456,
+      "required_pages_per_second_upper_estimate": 0.0021627578363195895
     },
     {
       "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": 28275.235835919706,
-      "required_pages_per_second_upper_estimate": 0.3510837979552219
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28094.05127316456,
+      "required_pages_per_second_upper_estimate": 0.3488340920715269
     }
   ],
   "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": 0.0945624471096096,
-    "producer_p95_seconds": 0.29891200799829676,
-    "gpu_update_seconds": 0.9778167779197481,
-    "producer_batches_per_second": 10.575022438250526,
-    "gpu_batches_per_second": 1.0226864813339014,
-    "producer_headroom": 10.340434366999167,
+    "producer_batch_seconds": 0.10508797940642012,
+    "producer_p95_seconds": 0.3139380370012077,
+    "gpu_update_seconds": 0.9841229280594846,
+    "producer_batches_per_second": 9.51583621312741,
+    "gpu_batches_per_second": 1.0161332202388802,
+    "producer_headroom": 9.364752596997425,
     "sustainable": true,
-    "buffer_seconds": 60.62464023102438,
+    "buffer_seconds": 60.031498611628564,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355958.19272034994,
-    "upload_bytes_per_second": 5580444.120847845,
+    "generation_bytes_per_second": 473489.57469207817,
+    "upload_bytes_per_second": 5482032.493952458,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.38034187515139,
+    "estimated_full_shards_per_hour": 101.59984045573958,
     "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": 1.5031630230005248,
-      "bytes": 11828419,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1863015779999841,
       "bytes": 6891470,
@@ -562,6 +556,12 @@
       "bytes": 6074322,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 0.9218759530012903,
+      "bytes": 1088353,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 258300, "loss": 0.596388328820467, "lr": 0.0003, "grad_norm": 0.14077683984788145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37807949727, "tokens_per_second": 146497.8689961445, "data_wait_s": 9.57899828790687e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 258400, "loss": 0.6657634973526001, "lr": 0.0003, "grad_norm": 0.18962240788548118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37822479157, "tokens_per_second": 147833.52644985594, "data_wait_s": 1.2987999070901424e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 258500, "loss": 0.6101230606436729, "lr": 0.0003, "grad_norm": 0.14810045976350694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37837180088, "tokens_per_second": 149825.3454071128, "data_wait_s": 1.1959997209487483e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 258600, "loss": 0.5844476539641619, "lr": 0.0003, "grad_norm": 0.15140719465755104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37851864597, "tokens_per_second": 148721.8763280229, "data_wait_s": 9.61499972618185e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:215b525f3b2012fa40d653eb6e2b6e9f054af7805b386a95fd3a22ea4990efb3
-size 456549
+oid sha256:9369df69f484c97b3e4fbe065d3ed0a00e9432858080b6c6492963d3c906036c
+size 458667