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": 76880.41137189965,
-      "required_pages_per_second_upper_estimate": 0.005318765245839351
+      "required_token_positions_per_second": 77274.90258707396,
+      "required_pages_per_second_upper_estimate": 0.005346057063450839
     },
     {
       "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": 9610.051421487457,
-      "required_pages_per_second_upper_estimate": 0.004023527849482955
+      "required_token_positions_per_second": 9659.362823384245,
+      "required_pages_per_second_upper_estimate": 0.004044173503717973
     },
     {
       "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": 19220.102842974913,
-      "required_pages_per_second_upper_estimate": 0.004003040553994295
+      "required_token_positions_per_second": 19318.72564676849,
+      "required_pages_per_second_upper_estimate": 0.004023581083166261
     },
     {
       "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": 28830.15426446237,
-      "required_pages_per_second_upper_estimate": 0.0017283237270578613
+      "required_token_positions_per_second": 28978.088470152732,
+      "required_pages_per_second_upper_estimate": 0.0017371921568065456
     },
     {
       "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": 28830.15426446237,
-      "required_pages_per_second_upper_estimate": 0.002241417683615081
+      "required_token_positions_per_second": 28978.088470152732,
+      "required_pages_per_second_upper_estimate": 0.002252918917413703
     },
     {
       "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": 28830.15426446237,
-      "required_pages_per_second_upper_estimate": 0.3698022634966505
+      "required_token_positions_per_second": 28978.088470152732,
+      "required_pages_per_second_upper_estimate": 0.3716998046478718
     }
   ],
   "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": 0.11581305874250347,
-    "producer_p95_seconds": 0.3462997370006633,
-    "gpu_update_seconds": 0.9589959091575324,
-    "producer_batches_per_second": 8.634604861126936,
-    "gpu_batches_per_second": 1.0427573156995937,
-    "producer_headroom": 8.280550739012476,
+    "producer_batch_seconds": 0.2396509237185569,
+    "producer_p95_seconds": 0.3698820439967676,
+    "gpu_update_seconds": 0.9541001998277866,
+    "producer_batches_per_second": 4.172735846302799,
+    "gpu_batches_per_second": 1.0481079452456863,
+    "producer_headroom": 3.981208104786069,
     "sustainable": true,
-    "buffer_seconds": 60.416742276924545,
+    "buffer_seconds": 60.10831258915056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397003.27887245466,
-    "upload_bytes_per_second": 4639770.208968742,
+    "generation_bytes_per_second": 386352.43781996047,
+    "upload_bytes_per_second": 4678812.920623213,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.1876618826888,
+    "estimated_full_shards_per_hour": 82.90223933171377,
     "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": 1.7626394689978042,
-      "bytes": 7617654,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.338195986001665,
       "bytes": 5044470,
@@ -562,6 +556,12 @@
       "bytes": 6154943,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.8716023119995953,
+      "bytes": 11724100,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -386,3 +386,4 @@
 {"step": 643000, "loss": 0.5572802424430847, "lr": 0.0001, "grad_norm": 0.1743637421779304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94118435350, "tokens_per_second": 150919.23249322307, "data_wait_s": 1.2418000551406294e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 643100, "loss": 0.5276366472244263, "lr": 0.0001, "grad_norm": 0.1811679444098251, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94133046313, "tokens_per_second": 151640.38554845308, "data_wait_s": 2.1270003344397992e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 643200, "loss": 0.5231402516365051, "lr": 0.0001, "grad_norm": 0.16027430736682807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94147697528, "tokens_per_second": 151841.47386547425, "data_wait_s": 1.4810997527092695e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 643300, "loss": 0.5461611747741699, "lr": 0.0001, "grad_norm": 0.17731107872806595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94162422000, "tokens_per_second": 152272.72209917265, "data_wait_s": 6.2920007621869445e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6297fa0021f0f35fb21eed7fee9c746849b1db2c5d4092c4051e546f2edc470e
-size 828213
+oid sha256:76f7b0eaaae9273cc68b094add9b127b65fb83bf405a3c9ea9fae289f4b1f81d
+size 830331