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": 75309.9508666543,
-      "required_pages_per_second_upper_estimate": 0.006527675669796154
+      "required_token_positions_per_second": 75883.0917392735,
+      "required_pages_per_second_upper_estimate": 0.006577354068022525
     },
     {
       "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": 9413.743858331787,
-      "required_pages_per_second_upper_estimate": 0.0037552831731018778
+      "required_token_positions_per_second": 9485.386467409187,
+      "required_pages_per_second_upper_estimate": 0.0037838624810153133
     },
     {
       "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": 18827.487716663574,
-      "required_pages_per_second_upper_estimate": 0.004355407930783397
+      "required_token_positions_per_second": 18970.772934818375,
+      "required_pages_per_second_upper_estimate": 0.004388554444269798
     },
     {
       "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": 28241.231574995363,
-      "required_pages_per_second_upper_estimate": 0.001719742832626347
+      "required_token_positions_per_second": 28456.15940222756,
+      "required_pages_per_second_upper_estimate": 0.001732830809665627
     },
     {
       "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": 28241.231574995363,
-      "required_pages_per_second_upper_estimate": 0.002174088183375676
+      "required_token_positions_per_second": 28456.15940222756,
+      "required_pages_per_second_upper_estimate": 0.0021906339224743157
     },
     {
       "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": 28241.231574995363,
-      "required_pages_per_second_upper_estimate": 0.350661578839482
+      "required_token_positions_per_second": 28456.15940222756,
+      "required_pages_per_second_upper_estimate": 0.3533302631365405
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2180268150311946,
-    "producer_p95_seconds": 0.2564067660005094,
-    "gpu_update_seconds": 0.9789941322700456,
-    "producer_batches_per_second": 4.586591790816755,
-    "gpu_batches_per_second": 1.0214565818502375,
-    "producer_headroom": 4.490246450327563,
+    "producer_batch_seconds": 0.07985358842954327,
+    "producer_p95_seconds": 0.19349934499950905,
+    "gpu_update_seconds": 0.9715998427333697,
+    "producer_batches_per_second": 12.522918752515723,
+    "gpu_batches_per_second": 1.0292303024532539,
+    "producer_headroom": 12.167265890507045,
     "sustainable": true,
-    "buffer_seconds": 60.69763620074283,
+    "buffer_seconds": 60.239190249468926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421984.8577451799,
-    "upload_bytes_per_second": 3518993.376063529,
+    "generation_bytes_per_second": 407059.9277426636,
+    "upload_bytes_per_second": 3593475.7487764894,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.5481271673827,
+    "estimated_full_shards_per_hour": 87.3455846234315,
     "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": 0.9026169319995461,
-      "bytes": 2322454,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.0145137629997407,
       "bytes": 6320844,
@@ -562,6 +556,12 @@
       "bytes": 5962047,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.150664095999673,
+      "bytes": 11982825,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -116,3 +116,4 @@
 {"step": 248700, "loss": 0.6067241448909044, "lr": 0.0003, "grad_norm": 0.15892990905708212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36403427514, "tokens_per_second": 149005.14933590667, "data_wait_s": 9.526000212645158e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 248800, "loss": 0.6059872824698687, "lr": 0.0003, "grad_norm": 0.1425244683854724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36417892741, "tokens_per_second": 148374.87166677753, "data_wait_s": 1.344299926131498e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 248900, "loss": 0.5806682482361794, "lr": 0.0003, "grad_norm": 0.15996525742501994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36432411760, "tokens_per_second": 148240.40482222656, "data_wait_s": 1.6429999959655106e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 249000, "loss": 0.5690943282097578, "lr": 0.0003, "grad_norm": 0.13730764362207964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36446962241, "tokens_per_second": 148038.11439091066, "data_wait_s": 1.4449999071075581e-05, "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 sha256:26238af80c4ce6e1436d54fc90517c944f1726e33d37a4c5cf20070a00bcbc95
-size 251655
+oid sha256:17686e400a9a2abc6b149b01c895861bb818973392a13c9fc2634bc1531d3f64
+size 253773