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": 76817.58434651633,
-      "required_pages_per_second_upper_estimate": 0.00531441872644455
+      "required_token_positions_per_second": 77242.74450253251,
+      "required_pages_per_second_upper_estimate": 0.005343832292545226
     },
     {
       "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": 9602.198043314542,
-      "required_pages_per_second_upper_estimate": 0.004020239804039141
+      "required_token_positions_per_second": 9655.343062816564,
+      "required_pages_per_second_upper_estimate": 0.004042490513910431
     },
     {
       "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": 19204.396086629084,
-      "required_pages_per_second_upper_estimate": 0.003999769250862479
+      "required_token_positions_per_second": 19310.68612563313,
+      "required_pages_per_second_upper_estimate": 0.0040219066629301956
     },
     {
       "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": 28806.594129943627,
-      "required_pages_per_second_upper_estimate": 0.0017269113329677053
+      "required_token_positions_per_second": 28966.029188449695,
+      "required_pages_per_second_upper_estimate": 0.0017364692212819054
     },
     {
       "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": 28806.594129943627,
-      "required_pages_per_second_upper_estimate": 0.0022395859867863285
+      "required_token_positions_per_second": 28966.029188449695,
+      "required_pages_per_second_upper_estimate": 0.0022519813613044665
     },
     {
       "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": 28806.594129943627,
-      "required_pages_per_second_upper_estimate": 0.3695000593879456
+      "required_token_positions_per_second": 28966.029188449695,
+      "required_pages_per_second_upper_estimate": 0.3715451211304331
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06534190967187214,
-    "producer_p95_seconds": 0.2243331190002209,
-    "gpu_update_seconds": 0.9597802459840505,
-    "producer_batches_per_second": 15.304113470538372,
-    "gpu_batches_per_second": 1.0419051696304842,
-    "producer_headroom": 14.68858579132114,
+    "producer_batch_seconds": 0.2731050749843291,
+    "producer_p95_seconds": 0.2418229920003796,
+    "gpu_update_seconds": 0.9544974155803425,
+    "producer_batches_per_second": 3.6615943517614253,
+    "gpu_batches_per_second": 1.047671773309089,
+    "producer_headroom": 3.4949823456598597,
     "sustainable": true,
-    "buffer_seconds": 60.466155496995185,
+    "buffer_seconds": 60.13333718156158,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384581.0099963201,
+    "generation_bytes_per_second": 371544.46961649065,
     "upload_bytes_per_second": 5140360.565549419,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.5221321574898,
+    "estimated_full_shards_per_hour": 79.72479406710663,
     "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": [

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

@@ -191,3 +191,4 @@
 {"step": 623500, "loss": 0.5607584118843079, "lr": 0.0001, "grad_norm": 0.17825670277705963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91263664963, "tokens_per_second": 152793.54941587502, "data_wait_s": 1.3788001524517313e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 623600, "loss": 0.5415359735488892, "lr": 0.0001, "grad_norm": 0.15616119691441457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91278380949, "tokens_per_second": 152902.4984811512, "data_wait_s": 1.572499968460761e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 623700, "loss": 0.5462983846664429, "lr": 0.0001, "grad_norm": 0.17555965936058146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91293035014, "tokens_per_second": 151753.54044312995, "data_wait_s": 1.3606000720756128e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 623800, "loss": 0.5751963257789612, "lr": 0.0001, "grad_norm": 0.15222753362728927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91307770224, "tokens_per_second": 152999.87859197438, "data_wait_s": 1.3946999388281256e-05, "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:a3daf95cbcdfc109175e1ebc297bb9ac990044244038edf1cf270bcfc4f353fa
-size 412071
+oid sha256:4b0125794bf5c1bfa7ccbac9757335d3c13d0a217e0904083870850cd8bbd936
+size 414189