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": 76814.9946316647,
-      "required_pages_per_second_upper_estimate": 0.005314239563962147
+      "required_token_positions_per_second": 76789.85829493667,
+      "required_pages_per_second_upper_estimate": 0.005312500573862965
     },
     {
       "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": 9601.874328958087,
-      "required_pages_per_second_upper_estimate": 0.0040201042715980176
+      "required_token_positions_per_second": 9598.732286867084,
+      "required_pages_per_second_upper_estimate": 0.004018788764187814
     },
     {
       "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": 19203.748657916174,
-      "required_pages_per_second_upper_estimate": 0.003999634408535423
+      "required_token_positions_per_second": 19197.464573734167,
+      "required_pages_per_second_upper_estimate": 0.003998325599522755
     },
     {
       "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": 28805.622986874263,
-      "required_pages_per_second_upper_estimate": 0.0017268531144235464
+      "required_token_positions_per_second": 28796.19686060125,
+      "required_pages_per_second_upper_estimate": 0.0017262880325463407
     },
     {
       "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": 28805.622986874263,
-      "required_pages_per_second_upper_estimate": 0.0022395104846842924
+      "required_token_positions_per_second": 28796.19686060125,
+      "required_pages_per_second_upper_estimate": 0.0022387776448277133
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28805.622986874263,
-      "required_pages_per_second_upper_estimate": 0.369487602607384
+      "required_token_positions_per_second": 28796.19686060125,
+      "required_pages_per_second_upper_estimate": 0.3693666943805396
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09124824976564128,
-    "producer_p95_seconds": 0.26038329400034854,
-    "gpu_update_seconds": 0.9598126036919337,
-    "producer_batches_per_second": 10.95911431253053,
-    "gpu_batches_per_second": 1.0418700443747924,
-    "producer_headroom": 10.518696042467464,
+    "producer_batch_seconds": 0.08754691208568488,
+    "producer_p95_seconds": 0.25593282599948,
+    "gpu_update_seconds": 0.9601267880560921,
+    "producer_batches_per_second": 11.42244741906224,
+    "gpu_batches_per_second": 1.0415291109881817,
+    "producer_headroom": 10.966997752203826,
     "sustainable": true,
-    "buffer_seconds": 60.46819403259182,
+    "buffer_seconds": 60.4879876475338,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369900.9878377646,
-    "upload_bytes_per_second": 5305071.967418788,
+    "generation_bytes_per_second": 344653.14986987703,
+    "upload_bytes_per_second": 5373950.195730494,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.37214113568977,
+    "estimated_full_shards_per_hour": 73.95454284736856,
     "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": 0.9842067560020951,
-      "bytes": 1088907,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1375966550003795,
       "bytes": 6241738,
@@ -562,6 +556,12 @@
       "bytes": 11525749,
       "pages": 46,
       "shards": 4
+    },
+    {
+      "seconds": 1.0188665900022897,
+      "bytes": 6807685,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -252,3 +252,4 @@
 {"step": 745200, "loss": 0.5324106812477112, "lr": 5e-05, "grad_norm": 0.24198127499305994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109076959884, "tokens_per_second": 149831.06108983708, "data_wait_s": 1.5336001524701715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745300, "loss": 0.5240442156791687, "lr": 5e-05, "grad_norm": 0.23911306087253575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109091459873, "tokens_per_second": 149986.14113557548, "data_wait_s": 1.0012001439463347e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745400, "loss": 0.5447136759757996, "lr": 5e-05, "grad_norm": 0.22656630798232283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109105958134, "tokens_per_second": 147603.6770643994, "data_wait_s": 1.3957996998215094e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 745500, "loss": 0.5598964095115662, "lr": 5e-05, "grad_norm": 0.2301542085667018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109120644051, "tokens_per_second": 151447.52547444103, "data_wait_s": 1.556699862703681e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b31a3a33551f81d12a69e965837abe199301145d8b79e516ba8a5dffdb8265cd
-size 542052
+oid sha256:f3836f85809b6ffafeb5e981522ee069cddcbf1f4bb99182b35ca193e548c7eb
+size 544170