Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75769.74247110738,
-      "required_pages_per_second_upper_estimate": 0.005241926594175636
+      "required_token_positions_per_second": 77166.09774188043,
+      "required_pages_per_second_upper_estimate": 0.005338529691798327
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9471.217808888423,
-      "required_pages_per_second_upper_estimate": 0.003965401115063247
+      "required_token_positions_per_second": 9645.762217735053,
+      "required_pages_per_second_upper_estimate": 0.004038479214145599
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18942.435617776846,
-      "required_pages_per_second_upper_estimate": 0.0039452097935626885
+      "required_token_positions_per_second": 19291.524435470106,
+      "required_pages_per_second_upper_estimate": 0.004017915788196989
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28413.65342666527,
-      "required_pages_per_second_upper_estimate": 0.0017033551377918814
+      "required_token_positions_per_second": 28937.28665320516,
+      "required_pages_per_second_upper_estimate": 0.0017347461501813057
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28413.65342666527,
-      "required_pages_per_second_upper_estimate": 0.002209036575469931
+      "required_token_positions_per_second": 28937.28665320516,
+      "required_pages_per_second_upper_estimate": 0.0022497467556142556
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28413.65342666527,
-      "required_pages_per_second_upper_estimate": 0.36445983795314674
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28937.28665320516,
+      "required_pages_per_second_upper_estimate": 0.3711764427496461
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.6818963261014233,
-    "producer_p95_seconds": 0.23811211900101625,
-    "gpu_update_seconds": 0.9730533270337306,
-    "producer_batches_per_second": 1.4664985888943811,
-    "gpu_batches_per_second": 1.027692904610289,
-    "producer_headroom": 1.4269813310139488,
+    "producer_batch_seconds": 0.05805863463288574,
+    "producer_p95_seconds": 0.19231988499996078,
+    "gpu_update_seconds": 0.9554454891138746,
+    "producer_batches_per_second": 17.22396687974431,
+    "gpu_batches_per_second": 1.0466321850840987,
+    "producer_headroom": 16.45656145989848,
     "sustainable": true,
-    "buffer_seconds": 60.329306276091295,
+    "buffer_seconds": 60.1930658141741,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343500.07514453086,
-    "upload_bytes_per_second": 5582422.698415957,
+    "generation_bytes_per_second": 343757.5345896887,
+    "upload_bytes_per_second": 5601983.848156459,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.7071198535151,
+    "estimated_full_shards_per_hour": 73.76236465709682,
     "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": 1.6309672779998436,
-      "bytes": 11574403,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.1955943939992721,
       "bytes": 921426,
@@ -562,6 +556,12 @@
       "bytes": 6264701,
       "pages": 26,
       "shards": 3
+    },
+    {
+      "seconds": 1.4136529820007127,
+      "bytes": 11858220,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 773000, "loss": 0.5889947414398193, "lr": 3e-05, "grad_norm": 0.260347812918309, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113144927660, "tokens_per_second": 153581.3207197683, "data_wait_s": 1.0908001058851369e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 773100, "loss": 0.5191934704780579, "lr": 3e-05, "grad_norm": 0.24382179405174736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113159528151, "tokens_per_second": 120095.38229532991, "data_wait_s": 1.2694999895757064e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 773200, "loss": 0.5310789346694946, "lr": 3e-05, "grad_norm": 0.25695695842821425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113174036181, "tokens_per_second": 151391.51501430312, "data_wait_s": 7.618000381626189e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 773300, "loss": 0.5370212197303772, "lr": 3e-05, "grad_norm": 0.2433236307142802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113188694398, "tokens_per_second": 152259.72881415344, "data_wait_s": 1.0412999472464435e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5bcc3f15c2ec653d5dbea1d661f53fd68f0d2d395045e3ea60ea894efc1a64e8
-size 281307
+oid sha256:c4d43c660ced0a67dc3d0c86380e79580bc47b3d70ea5f00b944b9d665784301
+size 283425