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": 76578.36655994454,
-      "required_pages_per_second_upper_estimate": 0.0052978690849077754
+      "required_token_positions_per_second": 76906.45542467975,
+      "required_pages_per_second_upper_estimate": 0.005320567033841207
     },
     {
       "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": 9572.295819993067,
-      "required_pages_per_second_upper_estimate": 0.0040077203675638684
+      "required_token_positions_per_second": 9613.306928084969,
+      "required_pages_per_second_upper_estimate": 0.004024890862112625
     },
     {
       "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": 19144.591639986134,
-      "required_pages_per_second_upper_estimate": 0.0039873135617760685
+      "required_token_positions_per_second": 19226.613856169937,
+      "required_pages_per_second_upper_estimate": 0.004004396626335854
     },
     {
       "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": 28716.887459979203,
-      "required_pages_per_second_upper_estimate": 0.001721533555077492
+      "required_token_positions_per_second": 28839.920784254908,
+      "required_pages_per_second_upper_estimate": 0.0017289092150068119
     },
     {
       "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": 28716.887459979203,
-      "required_pages_per_second_upper_estimate": 0.0022326116877745355
+      "required_token_positions_per_second": 28839.920784254908,
+      "required_pages_per_second_upper_estimate": 0.0022421769875705703
     },
     {
       "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": 28716.887459979203,
-      "required_pages_per_second_upper_estimate": 0.3683493985451598
+      "required_token_positions_per_second": 28839.920784254908,
+      "required_pages_per_second_upper_estimate": 0.36992753792607724
     }
   ],
   "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": 0.09964710097636953,
-    "producer_p95_seconds": 0.3395600780058885,
-    "gpu_update_seconds": 0.9627784361564657,
-    "producer_batches_per_second": 10.035414881132784,
-    "gpu_batches_per_second": 1.0386605707457757,
-    "producer_headroom": 9.661881045438347,
+    "producer_batch_seconds": 0.2801769202109199,
+    "producer_p95_seconds": 0.3526628879990312,
+    "gpu_update_seconds": 0.9586711491625999,
+    "producer_batches_per_second": 3.5691733610576857,
+    "gpu_batches_per_second": 1.0431105607731086,
+    "producer_headroom": 3.421663527605711,
     "sustainable": true,
-    "buffer_seconds": 60.65504147785734,
+    "buffer_seconds": 60.396282397243795,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454172.71488623496,
-    "upload_bytes_per_second": 2881985.7215351244,
+    "generation_bytes_per_second": 433876.2463984563,
+    "upload_bytes_per_second": 2874412.1196986893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.45489201488768,
+    "estimated_full_shards_per_hour": 93.09974235501544,
     "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": 3.3732278300012695,
-      "bytes": 11426245,
-      "pages": 26,
-      "shards": 3
-    },
     {
       "seconds": 2.9584781199955614,
       "bytes": 1202687,
@@ -562,6 +556,12 @@
       "bytes": 5924226,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.9274514060016372,
+      "bytes": 6142296,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -435,3 +435,4 @@
 {"step": 886000, "loss": 0.54771488904953, "lr": 3e-05, "grad_norm": 0.31169686749996584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129685560151, "tokens_per_second": 151801.6015801023, "data_wait_s": 1.4566001482307911e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 886100, "loss": 0.4592820107936859, "lr": 3e-05, "grad_norm": 0.31501846194969857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129700265402, "tokens_per_second": 152859.72386339528, "data_wait_s": 1.7710997781250626e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 886200, "loss": 0.5351279377937317, "lr": 3e-05, "grad_norm": 0.3124352346805511, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129714919216, "tokens_per_second": 152400.64649286724, "data_wait_s": 1.4372999430634081e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 886300, "loss": 0.5460257530212402, "lr": 3e-05, "grad_norm": 0.3166178739398988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129729689030, "tokens_per_second": 153508.47253516948, "data_wait_s": 1.2065000191796571e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3dcbcef8928077e278a4f98411051efbc04293c040950ca7dcd724f30eb95c81
-size 932778
+oid sha256:f67ea3cd439660df2fe77949d3d9d45df6cbecf64d69680b12abe13f79b7e97a
+size 934896